Caregiver‐mediated exercises for improving outcomes after stroke

Summary of findings for the main comparison. Caregiver‐mediated exercises compared with control intervention for people with stroke

Caregiver‐mediated exercises compared with control intervention for people with stroke
Patient or population: people with stroke Settings: inpatient and outpatient settings Intervention: caregiver‐mediated exercises Comparison: control, i.e. usual care, other intervention, no intervention
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control intervention	Caregiver‐mediated intervention
Patient: ADL measures Barthel Index. Scale 0 to 100 (follow‐up: 2 studies; 3/6 months) FIM. Scale 7 to 126 (no follow‐up)	The mean Barthel Index score ranged across control groups from 78 to 84 1 study: The mean FIM score in the control group was 65	The mean Barthel Index score in the intervention groups was 5.09 higher (‐2.88 to 13.07 higher) 1 study: The mean FIM score in the intervention group was 11 higher (‐1.59 to 23.67 higher)	‐	Barthel Index: 247 (3) FIM: 48 (1) Total: 295	⊕⊕⊕⊝ Moderate	Higher scores are better More than half of the studies at low risk of bias (3 low risk of bias, 1 at unclear risk of bias) There was clinical heterogeneity SMD 0.21 (‐0.02 to 0.44)
Caregiver: measures of mood, burden and QoL: burden Caregiver Strain Index Scale. 0 to 13 (follow‐up 3 months) Caregiver Burden Scale. 22 to 88 (no follow‐up)	The mean Caregiver Strain Index score in the control group was 3.4 The mean Caregiver Burden Scale score in the control group was 46.6	The mean Caregiver Strain Index score in the intervention group was 0.50 higher (‐0.81 to 1.81 higher) The mean Caregiver Burden Scale score in the intervention group was 1.30 lower (‐4.88 to 7.48 lower)	‐	Caregiver Strain Index: 40 (1) Caregiver Burden Scale: 51 (1) Total: 91	⊕⊕⊕⊝ Moderate	Lower scores are better Both studies at low risk of bias Small total number of participants SMD ‐0.04 (‐0.45 to 0.37)
Gait and gait‐related measures: walking speed in m/s (follow‐up: 1 study, 9 months)	The mean walking speed ranged across control groups from 0.26 m/s to 0.46 m/s	The mean walking speed in the intervention group was 0.08 m/s higher (‐0.03 to 0.18)	‐	71 (2)	⊕⊝⊝⊝ Very low
Gait and gait‐related measures: walking distance measured with the Six‐Minute Walk Test in metres walked in 6 minutes (follow‐up: 1 study, 3 months)	The mean distance walked ranged across control groups from 157 m to 166 m	The mean distance walked in the intervention groups was 30.98 m higher (‐20.22 to 82.19 higher)	‐	91 (2)	⊕⊕⊕⊝ moderate	Lower scores are better 1 study at unclear risk of bias Small total number of participants MD 0.04 (‐0.10 to 0.18)
Measures of mood and QoL of the patient: Stroke Impact Scale Stroke Impact Scale mobility scale. Scale 9 to 45. (no follow‐up)	The mean Stroke Impact Scale mobility score in the control group was 66.8	The mean Stroke Impact Scale mobility score in the intervention group was 18.2 higher (7.54 to 28.86 higher)	‐	51 (1)	⊕⊝⊝⊝ Very low	Higher scores are better 1 study at low risk of bias Small total number of participants MD 18.2 (7.54 to 28.86)
Length of stay: length of stay in rehabilitation unit in days	The mean length of stay in a rehabilitation unit in the control group was 52.3 days	The mean length of stay in a rehabilitation unit in the intervention group was 12 days lower (‐10.88 to 34.88)	‐	20 (1)	⊕⊝⊝⊝ very low	Higher scores are better 1 study at low risk of bias and 1 at unclear or high risk of bias Small total number of participants There was clinical heterogeneity MD 0.08 m/s (‐0.03 to 0.18)
Adverse outcomes: falls number of falls/patient (no follow‐up)	1 study: the mean number of falls/patient in the control group was 0.08	1 study: the mean number of falls/ patient in the intervention group was 0.04 lower (‐0.10 to 0.18 lower)	‐	48 (1)	⊕⊝⊝⊝ Very low	Higher scores are better Both studies at low risk of bias Small total number of participants MD 30.98 m (‐20.22 to 82.19)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). ADL: activities of daily living; CI: confidence interval; FIM: Functional Independence Measure; MD: mean difference; QoL: quality of life; RR: risk ratio; SD: standard deviation; SMD: standardised mean difference.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Antecedentes

available in

Descripción de la afección

El accidente cerebrovascular es una causa importante de discapacidad a largo plazo en los adultos, con efectos sobre las actividades cotidianas (AC) y la calidad de vida (CdV). Aunque la mayoría de los pacientes dejan el contexto de rehabilitación con algún nivel de deambulación independiente, muchos presentan discapacidades residuales para caminar y se ha informado que después de la rehabilitación sólo el 7% de los supervivientes de un accidente cerebrovascular pueden caminar a un nivel que les permita la participación comunitaria (Ada 2009). A los 12 meses después del accidente cerebrovascular alrededor del 28% de los pacientes con accidente cerebrovascular todavía es dependiente para las AC básicas como vestirse, bañarse y la movilidad dentro del domicilio (Ullberg 2015). Pettersen y colegas informaron que el 32% de los pacientes con un accidente cerebrovascular que viven en el domicilio después de tres años todavía no participaban en AC ampliadas (Pettersen 2002). Cualquier tratamiento que mejore el resultado funcional puede reducir potencialmente la carga de esta enfermedad para el paciente, los cuidadores y la sociedad.

Descripción de la intervención

Varias revisiones sistemáticas han indicado que una mayor intensidad de entrenamiento en cuanto al tiempo pasado en la terapia con ejercicios puede dar lugar a un mejor resultado funcional en los pacientes con accidente cerebrovascular en cuanto a las AC y el rendimiento funcional (French 2010; Galvin 2008a; Kwakkel 2004; Kwakkel 2006; Langhorne 2011; Lohse 2014; Veerbeek 2011; Veerbeek 2014). Un método eficaz desde el punto de vista de los recursos para aumentar la intensidad del entrenamiento podría ser incluir a los cuidadores en el entrenamiento con ejercicios (De Weerdt 2002). Los ejercicios mediados por el cuidador (EMC) se definieron como los ejercicios en los que el paciente con accidente cerebrovascular realiza los ejercicios junto con el cuidador bajo el auspicio de un terapeuta físico u ocupacional. "Bajo el auspicio" significa que el terapeuta participa como un entrenador al adiestrar al paciente y al cuidador en cómo realizar los ejercicios y los evalúa de forma sistemática. De esta forma los ejercicios tienen como finalidad mejorar las AC que incluyen la movilidad como trasladarse, pararse y caminar.

De qué manera podría funcionar la intervención

La realización de los ejercicios junto con el cuidador tiene la posibilidad de aumentar la intensidad de la práctica sin aumentar los costos de la asistencia sanitaria. Lo anterior podría mejorar los resultados en cuanto a las funciones corporales, las actividades y la participación, así como el costo‐efectividad en los pacientes con accidente cerebrovascular.

Además, los cuidadores participan más activamente en los EMC que en los servicios de rehabilitación aplicados generalmente, lo que puede aumentar la sensación de empoderamiento con niveles reducidos de carga para el cuidador (Brereton 2002; Smith 2004a). Los EMC podrían dar lugar a una reducción en la duración de la estancia hospitalaria o del tratamiento ambulatorio en contextos hospitalarios, de rehabilitación y de enfermería y pueden mejorar los resultados en el autocuidado, el empoderamiento y la CdV de los pacientes y los cuidadores.

Por qué es importante realizar esta revisión

Varias revisiones sistemáticas han indicado que la terapia con ejercicios adicionales y el entrenamiento con tareas repetitivas tienen un efecto favorable significativo sobre el resultado funcional después del accidente cerebrovascular, y concluyen que mientras mayor sea el tiempo invertido en la terapia con ejercicios (Galvin 2008a; Kwakkel 2004; Kwakkel 2006; Lohse 2014; Veerbeek 2011), y un mayor número de repeticiones, mejor será el resultado (French 2010; Langhorne 2011; Veerbeek 2014). Por lo tanto, las guías clínicas recomiendan que los pacientes que están en un contexto de rehabilitación deben tener la oportunidad de adiestrarse intensamente (ESO 2008; NICE 2013; SIGN 2010; Veerbeek 2014). Por ejemplo, la guía para el accidente cerebrovascular en el Reino Unido recomienda una dosis diaria de 45 minutos de terapia con ejercicios (NICE 2013).

Actualmente los recursos en contextos de hospitalización no son suficientes para satisfacer estas recomendaciones. La mayoría de los pacientes ingresados en unidades de accidente cerebrovascular, salas de rehabilitación y residencias geriátricas pasan inactivos la mayor parte de su tiempo de caminata durante la semana de trabajo (Bernhardt 2004; Smith 2008; West 2012), y los fines de semana, los servicios de rehabilitación (que incluyen la terapia con ejercicios) en la mayoría de los contextos hospitalarios y de rehabilitación no están disponibles (Otterman 2012). Por lo tanto, es importante encontrar métodos innovadores como los EMC para mejorar la intensidad de entrenamiento después del accidente cerebrovascular, sin aumentar los costos.

Sin embargo, al cuidador asumir el papel de un terapeuta (en lugar de una función familiar) puede añadir a su carga otra tarea (Gordon 2004). Por lo tanto, es importante estudiar el estado de ánimo, la carga y la CdV de los cuidadores cuando se les incluye en los EMC de forma sistemática. Todavía no se ha realizado una revisión sistemática para evaluar el efecto de la participación del cuidador en el entrenamiento con ejercicios sobre el resultado funcional después del accidente cerebrovascular, ni para evaluar el efecto sobre el estado de ánimo y la carga para el cuidador cuando participa en los EMC.

Objetivos

available in

Métodos

available in

Criterios de inclusión de estudios para esta revisión

Tipos de estudios

Se incluyeron los ensayos controlados aleatorios (ECA), incluidos los ECA grupales. Un grupo del ensayo debe haber recibido EMC y en esta revisión este grupo se consideró el grupo experimental. El otro grupo (control) podría haber recibido tratamiento habitual, ningún tratamiento u otro tipo de intervención de rehabilitación o atención‐control, siempre que no fuera mediado por el cuidador. El tratamiento habitual se aceptó cuando se describió como atención habitual en el contexto del participante.

Tipos de participantes

Personas con al menos 18 años de edad que habían tenido un accidente cerebrovascular. El accidente cerebrovascular se define por la Organización Mundial de la Salud como "un síndrome clínico caracterizado por signos de trastorno focal o general de las funciones cerebrales que se desarrollan rápidamente, que persisten por más de 24 horas o provocan la muerte, sin otras causas evidentes diferentes de la vascular" (WHO 1989). Los ECA se incluyeron independientemente del tiempo transcurrido desde el accidente cerebrovascular y del contexto.

Tipos de intervenciones

Un grupo de ECA debe haber incluido EMC, mientras que en el otro grupo de ECA no se precisó de forma explícita la participación del cuidador. Se incluyeron los ensayos en los que el paciente y el cuidador se adiestraron o se instruyeron juntos, así como los ensayos en los que el cuidador se adiestró o instruyó solo. No hubo límites en el número de sesiones ni la frecuencia de administración. Se incluyeron todos los tipos de ejercicios siempre que tuvieran como objetivo mejorar las capacidades de los pacientes de realizar las actividades cotidianas. Por lo tanto, se excluyeron los ECA de intervenciones en el habla, la deglución o cognitivas realizadas junto con el cuidador. El cuidador se definió como una persona que no recibe retribución o la recibe de forma parcial y que ayuda voluntariamente a un paciente impedido a realizar las AC. En otras palabras, los servicios mediados no fueron aplicados por un profesional sanitario sino que, en la mayoría de los casos, alguien cercano al paciente y de forma voluntaria ofreció los servicios. Esta persona puede haber sido la pareja, un miembro de la familia o un amigo, pero también puede haber sido un voluntario. Se señaló que esta persona "no fuera un profesional", como un "asistente de terapia". Cuando un profesional sanitario proporcionó los ejercicios mediados, el ECA se excluyó. Se incluyeron las intervenciones proporcionadas en cualquier ámbito, por ejemplo en el domicilio, en el hospital o en un contexto de rehabilitación. Debido a que el cuidador puede ser el que proporcione la intervención, no se excluyeron los ensayos que combinaron los EMC con una intervención existente. Sin embargo, se diferenció entre los ensayos en los cuales los EMC fueron la única intervención (centrados en los EMC) y los ensayos en los que se utilizó al cuidador para proporcionar otra intervención existente. Se estableció contacto con los autores del ensayo cuando no estuvo claro si un ensayo cumplió esta definición.

Tipos de medida de resultado

Resultados primarios

Paciente: medidas de las AC básicas, como el Barthel index (BI) (Collin 1988; Mahoney 1965), Functional Independence Measure (FIM) (Dodds 1993), Rankin Scale modificada (mRS) (De Haan 1995; Dromerick 2003); medidas de las AC ampliadas, como el Nottingham Extended Activities of Daily Living (NEADL) Index (Nouri 1987), o el Frenchay Activities Index (FAI) (Wade 1985). De hallarlas, las escalas se combinaron bajo el mismo concepto.
Cuidador: medidas de la carga, por ejemplo el Caregiver Strain Index (CSI) (Robinson 1983). De hallarlas, las escalas se combinaron bajo el mismo concepto.

Cuando fue posible se distinguió entre los cuidadores que eran familia o amigos y otros tipos de cuidadores, como los voluntarios, para las medidas de resultado mencionadas anteriormente.

Resultados secundarios

Medidas de deterioro motor: Motricity Index (MI) (Collin 1990), Fugl‐Meyer Assessment (FMA) (Duncan 1983; Sanford 1993; Shelton 2001).
Medidas de la marcha y relacionadas con la marcha: velocidad de marcha, distancia de caminata, 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; Mao 2002; Stevenson 2001).
Medidas de actividades o función del miembro superior, por ejemplo, Action Research Arm Test (ARAT) (Chen 2012; Hsieh 1998; Platz 2005).
Medidas del estado de ánimo y CdV del paciente, por ejemplo, medidas en la Stroke Impact Scale (SIS) (Duncan 1999; Duncan 2002; Duncan 2003), y la Hospital Anxiety and Depression Scale (HADS) (Aben 2002; Bjelland 2002; Herrmann 1997; Zigmond 1983).
Medidas de fatiga del participante, por ejemplo, medidas en la Fatigue Severity Scale (FSS) (Valko 2008).
Duración de la estancia hospitalaria, en el centro de rehabilitación o en la residencia geriátrica, o del tratamiento en un consultorio de pacientes ambulatorios.
Resultados adversos, por ejemplo dolor, lesión o caídas. Cuando fue posible se comparó el número total de caídas entre los grupos y el número de pacientes que tuvieron al menos una caída entre los grupos.
Cuidador: medidas del estado de ánimo y CdV, por ejemplo, la HADS (Aben 2002; Bjelland 2002; Herrmann 1997; Zigmond 1983), o la CarerQoL (Brouwer 2006; Hoefman 2011).

Cuando se encontraron escalas de medición de la misma construcción, se combinaron. Cuando los estudios informaron sobre medidas de resultado diferentes de las mencionado anteriormente, se comprobó si midieron la misma construcción. Si este fue el caso, se agruparon; si no midieron la misma construcción, los resultados se informaron por separado.

Results

Description of studies

See Characteristics of included studies, Characteristics of excluded studies, and Characteristics of ongoing studies tables.

Results of the search

Through electronic searches we found 8107 citations. In addition, one potentially relevant trial was already known to us, but not found through electronic searches (Wall 1987). After removing duplicates, we screened 5640 citations. Based on screening of titles, we excluded 5201 obviously irrelevant studies and screened the remaining 439 abstracts. Subsequently, we excluded 307 studies based on the abstract. Finally, we assessed 132 full‐text articles or trial registry entries for eligibility. After an extensive search, we still could not obtain full‐text articles for four studies ("THE DAYS AFTER"; "Family boosts results of poststroke therapy"; Liu 2012; Wang 2014). We identified 11 relevant systematic reviews, which we screened for trials (Bakas 2014; Brereton 2007; Glasdam 2010; Klinke 2015; Lawler 2013; Legg 2011; Morris 2014; Parke 2015; Pollock 2014a; Pollock 2014b; Warner 2015). In total, we identified 46 potentially relevant trials. The results of the search are summarised in Figure 1. We were able to include nine trials for final analysis (see Characteristics of included studies table), and we included six trials in the meta‐analysis (Abu Tariah 2010; Barzel 2015; Dai 2013; Galvin 2011; Wall 1987; Wang 2015).

Figure 1

Study flow diagram.

We excluded three trials from the meta‐analysis because of poor methodological quality (Agrawal 2013; Gómez 2014) or no reporting of required data (i.e. means or SDs, or both, of outcome measures) (Agrawal 2013; Gómez 2014; Souza 2015), or both. We had no success contacting the corresponding authors to request the necessary data.

We excluded 37 trials, 35 with reasons given in the Characteristics of excluded studies table. Two trials are ongoing (see Characteristics of ongoing studies table).

Included studies

Participants

Characteristics

In the nine included studies, 456 stroke survivors and their caregivers were randomised to CME or control interventions. A total of 342 people with stroke‐caregiver couples were included in the six trials included in the meta‐analysis. In these six trials, nine patient‐caregiver couples were not analysed according to intention‐to‐treat principles and no information about these withdrawals was published. Therefore, we have presented information about 333 stroke survivors and their caregivers (ranging from 18 to 156 patient‐caregiver couples per trial) in the meta‐analysis.

The mean age in all studies was around 60 years. The mean time since onset of symptoms ranged from 15 days to 10 years. One trial did not report mean time since onset of symptoms (Gómez 2014).

Three studies defined inclusion or exclusion criteria for the caregiver, for example "willing to participate", "medically stable and physically able" (Galvin 2011), "being defined as primary caregivers" (Dai 2013), and "caregivers were excluded if they were in poor physical health, had mental or behavioural disorders" (Wang 2015).

Four studies described an inclusion criterion for the patient about the caregiver: "live with family caregiver at home" (Abu Tariah 2010), "patients with family support" (Gómez 2014), "had a caregiver who was prepared to be a non‐professional coach (e.g., family member)" (Barzel 2015), and "availability of a family member to supervise home exercises" (Souza 2015). Two studies gave information about the caregiver: "about 50% of the caregivers were nursing attendants" (Dai 2013), and "majority were patients' spouse" (Wang 2015).

Sample size

Five trials included fewer than 50 participants: 20 participants (Abu Tariah 2010; Wall 1987), 24 participants (Souza 2015), 30 participants (Agrawal 2013), and 40 participants (Galvin 2011). Four trials included more than 50 participants: 51 participants (Wang 2015), 55 participants (Dai 2013), 60 participants (Gómez 2014), and 156 participants (Barzel 2015).

Interventions

The content of the training and the timing was different between trials. Details of each intervention are summarised in Table 1.

See: Summary of findings for the main comparison Caregiver‐mediated exercises compared with control intervention for people with stroke

Table 1. Outline of included studies

Study ID	Form of training	Upper or lower body	Timing since stroke	Task caregiver	Routine care continued	Control group	Programme (length ‐frequency‐ duration)	Contact with therapist	Place
Abu Tariah 2010	CIMT	Upper	> 2 months	Carried out the intervention with support of therapists	No	Neurodevelopmental training, same intensity	2 months ‐ daily ‐ 2 hours	3 or 4 sessions	Home
Agrawal 2013	Exercise therapy	Upper	"Sub‐acute stroke"	Encouragement, participating, and help	Yes	Usual care	4 weeks ‐5 days/week ‐ 60 to 90 minutes	Weekly	Inpatient?
Barzel 2015	CIMT	Upper	> 6 months	Supervision, help, and maintaining training diary	No	Usual care, frequency of seeing a therapist was the same	4 weeks ‐ Every weekday (not weekend) ‐ 2 hours	5 x 60 minutes	Home
Dai 2013	Vestibular rehabilitation	Both	< 6 months	Guidance and supervision (in third and fourth week)	Yes	Usual care	4 weeks ‐ 10 sessions per 2 weeks ‐ 30 minutes	2 to 4 sessions in first 2 weeks	Inpatient?
Galvin 2011	Exercise therapy	Lower	Assessment 2 weeks after stroke onset	Encouragement and help	Yes	Usual care	8 weeks ‐every day ‐ 35 minutes	Weekly	Inpatient or at home
Gómez 2014	CIMT	Upper	< 6 months	Monitoring and supervising	Yes	Usual care	14 days ‐ every day* ‐ 5.5 hours*	1.5 hours per day*	Inpatient
Souza 2015	CIMT: 1.5 hours with therapist and 1.5 hours with caregiver	Upper	< 24 months**	Supervision and making notes	No	CIMT: 3 hours with therapist	22 days ‐ 10 sessions ‐ 3 hours	10 x 90 minutes	Outpatient and home
Wall 1987	Exercise therapy	Lower	After discharge of rehabilitation	Supervision	No	No intervention	6 months ‐ twice a week ‐ 1 hour	1 group: twice a week 1 group: once a week 1 group: 'monitoring'	Outpatient or at home
Wang 2015	Exercise programme aimed at body functions, activities, and participation	Both	> 6 months	Encouragement and help	No	Usual care	12 weeks ‐ minimal twice a week, if possible every day ‐ minimal 50 to 60 minutes	Weekly 90 minutes	Home

CIMT: constraint‐induced movement therapy.
* Details of the intervention are not completely clear, contact with the authors was not successful.
** But mean time since stroke was 27 and 35 months since stroke, unclear why.

Two trials were aimed at the lower body (Galvin 2011; Wall 1987), five at the upper body (Abu Tariah 2010; Agrawal 2013; Barzel 2015; Gómez 2014; Souza 2015), and two at both upper and lower body (Dai 2013; Wang 2015). Four studies included patients within six months after stroke (Agrawal 2013; Dai 2013; Galvin 2011; Gómez 2014), three studies included patients beyond six months after stroke (Barzel 2015; Wall 1987; Wang 2015), one study included patients from two months after stroke or later (Abu Tariah 2010), one study included patients if they had a stroke in the last 24 months (Souza 2015). The task of the caregiver ranged across trials from supervision, guidance, encouragement, to physical help. In four trials, usual care continued, so CME were applied in addition to usual care (Agrawal 2013; Dai 2013; Galvin 2011; Gómez 2014). The frequency, duration, and programme length differed between studies, with training frequencies ranging from twice a week (Wall 1987; Wang 2015), to every day (Abu Tariah 2010; Galvin 2011), with a duration per session ranging from 30 minutes (Dai 2013), to three hours (Souza 2015), and a programme length ranging from 14 days (Gómez 2014), to six months (Wall 1987). In four trials, patients had weekly contact with the supervising therapist (Agrawal 2013; Barzel 2015; Galvin 2011; Wang 2015). Two trials planned two to four sessions with a therapist (Abu Tariah 2010; Dai 2013). One trial had 10 sessions with a therapist in 22 days (Souza 2015). One trial consisted of four groups, the amount of contact with the therapist differed between trial groups (Wall 1987). The frequency and duration of one trial was not clearly reported (Gómez 2014). Three trials were carried out at home (Abu Tariah 2010; Barzel 2015; Wang 2015), one trial was carried out in an inpatient setting (Gómez 2014), three trials were carried out when patients were inpatient, outpatient, or at home (Galvin 2011; Souza 2015; Wall 1987), and two trials were unclear about the location of the intervention (Agrawal 2013; Dai 2013).

Two trials had more than one trial group. The study by Agrawal 2013, which was not included in meta‐analysis, had two experimental trial groups with different duration of intervention (60 and 90 minutes, five days a week) and one control group. Wall 1987 had two intervention groups (CME, CME plus physiotherapy) and two control groups (physiotherapy, no intervention). We decided to combine the intervention groups and the control groups into one comparison because of the small total number of participants (20).

Compliance

Five studies recorded compliance: "frequency of training and tasks completed was recorded" (Wang 2015), "the amount of training was noted in a diary by patients' families" (Abu Tariah 2010), "compliance with therapy time was documented through the use of an exercise diary, in which the number of exercises completed and time taken to complete the exercises were recorded daily" (Galvin 2011), "a log sheet per participant to record the total number of minutes completed per day" (Agrawal 2013), and "compliance was assessed in all participants via a form (standard therapy group) or a training diary (home CIMT group)" (Barzel 2015). Two trials reported these outcomes in the results. Galvin 2011 reported that 245 minutes of additional exercise therapy was planned for each participant and that a mean of 227 minutes was actually delivered. Barzel 2015 reported a mean exercise time of 27.7 hours within the four‐week intervention. They also noted 12 cases of participants not adhering to the protocol. In Souza 2015, compliance about wearing of the sling was reported in the results, but no information about compliance to the CME was provided. Agrawal 2013 mentioned "inability to monitor patient's compliance with the home exercise programme which might have influenced the study".

Comparisons

Interventions consisted of CME in addition to usual care (Agrawal 2013; Dai 2013; Galvin 2011; Gómez 2014), or instead of usual care (Abu Tariah 2010; Barzel 2015; Souza 2015; Wall 1987; Wang 2015). Two studies included a control intervention (Abu Tariah 2010; Souza 2015), seven included usual care as control (Agrawal 2013; Barzel 2015; Dai 2013; Galvin 2011; Gómez 2014; Wall 1987; Wang 2015), one had no control intervention (Wall 1987). Furthermore, there were different forms of interventions in terms of type of exercise therapy, duration of the intervention, and timing of the intervention.

Outcome measures

All trials reported outcome measures at the end of intervention. Five trials reported outcome measures after three to six months' follow‐up (Abu Tariah 2010; Barzel 2015; Galvin 2011; Souza 2015; Wall 1987). Two trials reported outcome measures during the intervention period (Dai 2013; Wall 1987). Some outcome measures were not reported at baseline, but only at post intervention and at follow‐up. In some instances there were no SDs of outcome measures given, for which we imputed other SDs from the same study when possible (i.e. Galvin 2011: no SD at post intervention for NEADL Index, CSI and Reintegration to Normal Living Index was available and follow‐up SD was used; Abu Tariah 2010: no SD at post intervention or follow‐up for Wolf Motor Function test ‐ performance time was given and SD from baseline was used). Walking speed was reported in different units and were converted to metres/second. Where available, we also extracted mean changes from baseline (Abu Tariah 2010; Barzel 2015; Galvin 2011; Wang 2015), and in those cases where postintervention scores were not available, we used the mean change from baseline. Abu Tariah 2010 and Wang 2015 gave no SDs, but provided CIs. We calculated the SDs for these outcomes using the Z‐score.

One trial reported two outcome measures for extended ADL (Galvin 2011). Based on that, the NEADL Index is developed for people with stroke and widely used in stroke research, we restricted to NEADL Index in the main analysis.

Insufficient information was available regarding the type of caregiver, rendering it impossible to distinguish between caregivers who were family or friends and other (voluntary) caregivers for the different outcome measures. One study mentioned that "about 50% of the caregivers were nursing attendants" (Dai 2013), and one study included four paid workers (Wang 2015). We did not take this professional background into account during the analyses.

The trials used a variety of outcome measures. Some outcome measures were identical, but most differed between trials. We combined outcome measures when they appeared to measure the same construct.

Excluded studies

We excluded 35 articles based on the full texts because they did not meet the inclusion criteria (Adie 2014; Araujo 2015; Barzel 2009; Baskett 1999; Bertilsson 2014; Cameron 2015; Chang 2015; Chinchai 2010; El‐Senousey 2012; Evans 1984; Forster 2013; Goldberg 1997; Grasel 2005; Harrington 2010; Harris 2009; Hebel 2014; Hirano 2012; Jones 2015; Kalra 2004; Koh 2015; Larson 2005; Lin 2004; Maeshima 2003; Marsden 2010; McClellan 2004; Mudzi 2012; NCT00908479; Osawa 2010; Parker 2012; Redzuan 2012; Schure 2006; Shyu 2010; Smith 2004b; Van de Port 2012; Walker 1996). See Characteristics of excluded studies table.

The most common reasons for exclusion were: interventions were educational for patient and caregiver but they performed no, or minimal, exercises together (Chinchai 2010; El‐Senousey 2012; Evans 1984; Forster 2013; Harrington 2010; Larson 2005; Marsden 2010; Mudzi 2012; Parker 2012; Schure 2006; Shyu 2010; Smith 2004a); caregivers were involved and encouraged to participate but caregiver participation was not mandatory (Adie 2014; Baskett 1999; Bertilsson 2014; Harris 2009; Jones 2015; Lin 2004; McClellan 2004; NCT00908479; Van de Port 2012; Walker 1996); and the intervention concerned 'skill training' (Araujo 2015; Chang 2015; El‐Senousey 2012; Forster 2013; Grasel 2005; Hebel 2014; Kalra 2004; Mudzi 2012). Skill training is primarily aimed at training of the caregiver in performing ADL and mobility together with the patient to improve functioning together in the home situation. Skill training is given to the caregiver in a limited number of sessions by a professional, like a therapist or a nurse, but it is not considered progressive training to improve functioning of the patient.

Furthermore, there are some non‐randomised studies about CME (Barzel 2009; Hirano 2012; Maeshima 2003; Osawa 2010). Because of their relevance for the topic of this review they are listed in Characteristics of excluded studies table. However, it is important to note that our search was not aimed at identifying non‐randomised studies and, therefore, we may not be complete in reporting these studies.

Risk of bias in included studies

Assessments for 'Risk of bias' in individual studies are shown in the Characteristics of included studies table. See also Figure 2 and Figure 3 for a summary of the results.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

All trials used random allocation to an intervention or control group, of which four adequately described how the randomisation procedure took place and provided sufficient information to determine that the allocation procedure was concealed (Abu Tariah 2010; Barzel 2015; Galvin 2011; Wang 2015). One study was unclear about the randomisation procedure, but did provide sufficient information about allocation procedure (Souza 2015). The other four studies did not describe the randomisation procedure sufficiently (Agrawal 2013; Dai 2013; Gómez 2014; Wall 1987).

Blinding

Participant blinding

Due to the nature of the intervention, participants included in the trials could not be blinded for treatment allocation.

Investigator blinding

Six studies blinded the outcome assessors to treatment allocation (Abu Tariah 2010; Barzel 2015; Dai 2013; Galvin 2011; Souza 2015; Wang 2015). Three studies did not report anything about an outcome assessor (Agrawal 2013; Gómez 2014; Wall 1987). Five studies used participant‐reported outcomes (questionnaires, report of number of falls) (Barzel 2015; Dai 2013; Galvin 2011; Souza 2015; Wang 2015). For these outcomes, the assessor (patient or caregiver) was aware of the treatment allocation. This may have biased the results.

Incomplete outcome data

Three studies had no withdrawals and, therefore, reported complete outcome data (Agrawal 2013; Wall 1987; Wang 2015). Four studies had withdrawals, but reasons were well described and comparable in the intervention and control group (Barzel 2015; Dai 2013; Galvin 2011; Souza 2015). One study reported only withdrawals in the control group (Abu Tariah 2010). Reasons for withdrawal were not documented by the participants, making the risk of bias unclear. One trial did not describe withdrawals, making the risk of bias unclear (Gómez 2014).

Selective reporting

For two included trials (Barzel 2015; Galvin 2011), we identified a trial registry (NCT00666744) and published protocol (Barzel 2013; Galvin 2008b). Galvin 2011 reported no exclusion criteria in the trial paper in contrast to the protocol paper (Galvin 2008b) and trial registration (NCT00666744). Not all outcome measures that were reported in the protocol paper of Barzel 2013 were reported in the trial paper (Barzel 2015), such as the EQ‐5D and healthcare costs. There were an insufficient number of studies (fewer than 10) to reliably examine the effects of risk of bias on estimates of effect and thus we generated no funnel plots.

Other potential sources of bias

Three trials did not perform an intention‐to‐treat analysis. This could be a potential source of bias (Abu Tariah 2010; Dai 2013; Souza 2015). Three trials did not report means or SDs for (a part of) the study outcomes (Agrawal 2013; Galvin 2011; Souza 2015). In one trial, means and SDs for outcome measures were not given, the included outcomes were insufficiently described, and intervention and timing of measurements needed clarification (Gómez 2014). We identified no other potential sources of bias for the remaining trials (Barzel 2015; Wall 1987; Wang 2015).

Grading the quality of the evidence

We determined the quality of the evidence using GRADE levels of evidence. We downgraded effects based on one trial by two levels of evidence and effects based on a small total number of participants (fewer than 200 participants) (BMJ Clinical Evidence 2012) by one level. When half, or more, of the included trials for an outcome measure were of unclear or high risk of bias, we downgraded the level of evidence by one level. When we found substantial unexplained statistical heterogeneity or clinical heterogeneity, we also downgraded the level of evidence by one level. In addition, when we found publication bias, we downgraded the level of evidence by one level.

Effects of interventions

Caregiver‐mediated exercises versus control (Comparison 1 and 2): primary outcomes

Patient: activities of daily living measures

End of intervention

Three trials assessed the BI (100‐point version) (Barzel 2015; Galvin 2011; Wang 2015). We found no significant summary effect (mean difference (MD) 5.09, 95% CI ‐2.88 to 13.07; P = 0.21; Table 2). One trial used the FIM (Dai 2013). The effect of CME on the FIM was not significant (MD 11.04, 95% CI ‐1.59 to 23.67; P = 0.09; Table 2). Overall, we found no significant summary effect on basic ADL (standardised mean difference (SMD) 0.21, 95% CI ‐0.02 to 0.44; P =0.07; Analysis 1.1). The quality of evidence for effects on basic ADL was moderate; it was downgraded one level due to clinical heterogeneity between studies.

Table 2. (Standard) Mean differences which are not reported in section 'data and analysis'

Outcome	Outcome measure	Fixed‐effect or random‐effects model	Mean difference	Confidence interval	Heterogeneity	P value
1.1 Patient: ADL measures ‐ Combined	1.1.1 Barthel Index	Random‐effects	5.09	‐2.88 to 13.07	58%	0.21
1.1 Patient: ADL measures ‐ Combined	1.1.2 Functional Independence Measure	Fixed‐effect	11.04	‐1.59 to 23.67	‐	0.09
1.2 Patient: ADL measures ‐ extended ADL	1.2.1 Nottingham Extended Activities of Daily Living Index	Fixed‐effect	5.50	‐5.83 to 16.83	‐	0.34
1.2 Patient: ADL measures ‐ extended ADL	1.2.2 IADL	Fixed‐effect	0.02	‐0.72 to 0.76	‐	0.96
1.3 Caregiver: burden	1.3.1 Caregiver Strain Index	Fixed‐effect	‐0.50	‐1.81 to 0.81	‐	0.46
1.3 Caregiver: burden	1.3.2 Caregiver Burden Scale	Fixed‐effect	1.30	‐4.88 to 7.48	‐	0.68
1.6 Gait and gait‐related measures: balance	1.6.1 Berg Balance Scale	Fixed‐effect	6.35	1.64 to 11.06	0%	0.008
1.6 Gait and gait‐related measures: balance	1.6.2 Postural Assessment for Stroke patients	Fixed‐effect	3.50	‐0.52 to 7.52	‐	0.09
2.2 Patient: ADL measures ‐ extended ADL	2.2.1 Nottingham Extended Activities of Daily Living Index	Fixed‐effect	9.50	‐1.83 to 20.83	‐	0.10
2.2 Patient: ADL measures ‐ extended ADL	2.2.2 IADL	Fixed‐effect	0.02	‐0.77 to 0.81	‐	0.96
3.1 Patient: ADL measures ‐ combined	3.1.1 < 6 months	Fixed‐effect	0.44*	0.01 to 0.86	0%	0.04
3.1 Patient: ADL measures ‐ combined	3.1.2 > 6 months	Random‐effects	4.90	‐7.56 to 17.36	77%	0.44
8.1 Patient ADL measures ‐ extended ADL ‐ end of intervention	8.1.1 Reintegration to normal living Index	Fixed‐effect	0.20	‐3.76 to 4.16	‐	0.92
	8.1.2 IADL	Fixed‐effect	0.02	‐0.72 to 0.76	‐	0.96
8.2 Patient ADL measures ‐ extended ADL ‐ end of follow‐up	8.2.1 Reintegration to normal living Index	Fixed‐effect	4.50	0.54 to 8.46	‐	0.03
8.2 Patient ADL measures ‐ extended ADL ‐ end of follow‐up	8.2.2 IADL	Fixed‐effect	0.02	‐0.77 to 0.81	‐	0.96

ADL: activities of daily living; IADL: instrumental activities of daily living.

*Standardised mean difference.

Two trials assessed extended ADL (Barzel 2015; Galvin 2011). We found no significant effects of CME on the NEADL Index (MD 5.50, 95% CI ‐5.83 to 16.83; P = 0.34; Table 2) or on the Instrumental Activities of Daily Living (IADL) (MD 0.02, 95% CI ‐0.72 to 0.76; P = 0.96; Table 2). Overall, we found no significant summary effect on extended ADL (SMD 0.07, 95% CI ‐0.21 to 0.35; P = 0.64; Analysis 1.2). This effect was based on two trials with low risk of bias, but with clinical heterogeneity between studies and a small total number of participants for this outcome measure, resulting in a low quality of evidence.

Follow‐up

Two trials assessed basic ADL and extended ADL at three months' follow‐up (Galvin 2011) and six months' follow‐up (Barzel 2015). We found no significant summary effect of CME on basic ADL (MD 2.69, 95% CI ‐8.18 to 13.55; P = 0.63; Analysis 2.1). This effect was based on two trials with low risk of bias, but with clinical heterogeneity between studies and a small total number of participants for this outcome measure, resulting in a low quality of evidence. The substantial statistical heterogeneity between trials (I² =69%), can be explained by different timing post stroke (within six months versus beyond six months) and thus there was no reason to downgrade the level of evidence further.

The effect of CME on extended ADL measured with the NEADL Index (MD 9.50, 95% CI ‐1.83 to 20.83; P = 0.10; Table 2), or measured with the IADL (MD 0.02, 95% CI ‐0.77 to 0.81; P = 0.96; Table 2) was not significant. Overall, there was no significant summary effect of CME on extended ADL (SMD 0.11, 95% CI ‐0.17 to 0.39; P = 0.45; Analysis 2.2). The quality of evidence was low, based on two trials with low risk of bias, but with clinical heterogeneity between studies and a small total number of participants for this outcome measure.

Caregiver: measures of burden

End of intervention

One trial used the CSI to assess caregiver burden (Galvin 2011); we found no significant effect (MD –0.50, 95% CI ‐1.81 to 0.81; P = 0.46; Table 2). Another trial used the Caregiver Burden Scale (Wang 2015), and again we found no significant effect (MD 1.30, 95% CI ‐4.88 to 7.48; P = 0.68; Table 2). Overall, we found no significant summary effect of CME on caregiver strain (SMD ‐0.04, 95% CI ‐0.45 to 0.37; P = 0.86; Analysis 1.3). These findings were based on two trials with low risk of bias, but with a small total number of participants for this outcome measure, resulting in moderate quality of evidence.

Follow‐up

One study reported follow‐up of caregiver burden by using the CSI, three months after termination of the intervention (Galvin 2011). We found no significant effect of CME on caregiver strain compared with the control group (MD 0.60, 95% CI ‐0.71 to 1.91; P = 0.37; Analysis 2.3). The quality of the evidence for this finding was very low, since it is based on only one trial with a small number of participants.

Caregiver‐mediated exercises versus control (Comparison 1 and 2): secondary outcomes

Measures of motor impairment

One study assessed the FMA lower extremity score (Galvin 2011). We found no significant effect after the intervention (MD 3.10, 95% CI ‐2.02 to 8.22; P = 0.24; Analysis 1.4) or at follow‐up (MD 3.40, 95% CI ‐1.74 to 8.54; P = 0.19; Analysis 2.4). These findings were based on one trial with a small number of participants, resulting in a very low quality of evidence.

One study assessed the FMA upper extremity score (Abu Tariah 2010). We found no significant effect of CME at the end of intervention (MD 4.43, 95% CI ‐2.09 to 10.95; P = 0.18; Analysis 1.5) or at follow‐up (MD 2.75, 95% CI ‐8.24 to 13.74; P = 0.62; Analysis 2.5). These findings were based on only one trial with a small number of participants. Therefore, the quality of evidence was very low.

Gait and gait‐related measures

Balance

Two trials reported the BBS (Galvin 2011; Wang 2015). We found a significant summary effect (MD 6.35, 95% CI 1.64 to 11.06; P = 0.008; Table 2). One study assessed the Postural Assessment Scale for Stroke Patients (Dai 2013), and found no significant effect of CME (MD 3.50, 95% CI ‐0.52 to 7.52; P = 0.09; Table 2). Overall, we found a significant summary effect of CME on standing balance performance at the end of the intervention (SMD 0.53, 95% CI 0.19 to 0.87; P = 0.002; Analysis 1.6). These findings were based on a small total number of participants and there was clinical heterogeneity between studies resulting in a low quality of evidence. One trial was of unclear risk of bias (Dai 2013), but more than half of the trials were of low risk of bias (Galvin 2011; Wang 2015), and thus there was no reason to downgrade the level of evidence further.

Only one trial assessed standing balance performance at three months' follow‐up (Galvin 2011). There was no significant effect (MD 8.40, 95% CI ‐1.04 to 17.84; P = 0.08; Analysis 2.6). This effect was based on one trial with a small number of participants resulting in a very low quality of evidence.

Walking distance

Two trials used the Six‐Minute Walk Test to assess walking distance (Galvin 2011; Wang 2015). We found no significant summary effect of CME at the end of the intervention period (MD 30.98 m, 95% CI ‐20.22 to 82.19; P = 0.24; Analysis 1.7). These findings were based on two trials with a low risk of bias, but with a small total number of participants for this outcome measure, resulting in a moderate quality of evidence.

Only one trial assessed the Six‐Minute Walk Test at three months' follow‐up (Galvin 2011). There was a significant effect in favour of CME (MD 109.50 m, 95% CI 17.12 to 201.88; P = 0.02; Analysis 2.7). This finding was based on one trial with a small number of participants, resulting in a very low quality of evidence.

Walking speed

Two trials reported comfortable walking speed (Wall 1987; Wang 2015). We found no significant summary effect of CME on walking speed (MD 0.08 m/s, 95% CI ‐0.03 to 0.18; P = 0.17; Analysis 1.8). This effect was based on one trial with low risk of bias (Wang 2015) and one trial with an unclear risk of bias (Wall 1987). In addition, there was a small total number of participants. Therefore, the quality of evidence was low.

Only Wall 1987 reported follow‐up data three months after termination of the intervention. We found no significant effect of CME on walking speed (MD 0.10 m/s, 95% CI ‐0.02 to 0.22; P = 0.10; Analysis 2.8). The quality of evidence was very low, because the effect was based on only one trial of unclear risk of bias with a small total number of participants.

Measures of upper limb activities or function

Two trials with low risk of bias used the Wolf Motor Function test and the Motor Activity Log (Abu Tariah 2010; Barzel 2015). However, there may be publication bias, because all studies excluded for meta‐analysis were about upper limb training (Agrawal 2013; Gómez 2014; Souza 2015). In addition, there was a small total number of participants for these outcome measures and we detected substantial unexplained statistical heterogeneity between trials. We graded the quality of the evidence as very low, except the Wolf Motor Function test ‐ performance time and the Motor Activity Log ‐ amount of use at the end of intervention, and the Motor Activity Log ‐ quality of movement at both end of intervention and follow‐up. We did not detect any substantial statistical heterogeneity in these cases and, therefore, we graded the quality of evidence as low.

We found no significant summary effect of CME on the Wolf Motor Function test ‐ functional ability (end of intervention: MD 0.02, 95% CI ‐0.52 to 0.55; P = 0.95; Analysis 1.9; follow‐up four to six months after termination: MD 0.08, 95% CI ‐0.46 to 0.61; P = 0.77; Analysis 2.9), the Motor Activity Log ‐ amount of use (end of intervention: MD 0.01, 95% CI ‐0.36 to 0.38; P = 0.96; Analysis 1.11; follow‐up four to six months after termination: MD 0.21, 95% CI ‐0.65 to 1.08; P = 0.63; Analysis 2.11), and Motor Activity Log ‐ quality of movement (end of intervention: MD 0.08, 95% CI ‐0.26 to 0.42; P = 0.64; Analysis 1.12; follow‐up four to six months after termination: MD ‐0.03, 95% CI ‐0.43 to 0.37; P = 0.89; Analysis 2.12).

For the Wolf Motor Function test ‐ performance time, we found a significant summary effect in favour of the control group post intervention (MD ‐1.72, 95% CI ‐2.23 to ‐1.21; P < 0.00001; Analysis 1.10), but not at follow‐up (MD 1.85, 95% CI ‐8.78 to 12.48; P = 0.73; Analysis 2.10).

One trial used the Nine Hole Peg test (Barzel 2015). We found no significant effect post intervention (MD ‐0.04, 95% CI ‐0.11 to 0.03; P = 0.26; Analysis 1.13) or at follow‐up (MD ‐0.05, 95% CI ‐0.12 to 0.02; P = 0.17; Analysis 2.13). This evidence was based on one trial with a small number of participants, resulting in a very low quality of evidence.

Measures of mood and quality of life of the patient

One trial assessed QoL of the patients with the SIS 3.0 at the end of the intervention (Wang 2015), and one trial assessed only SIS hand function (Barzel 2015).

The effect of CME was significant for the composite physical scale (MD 12.40, 95% CI 1.67 to 23.13; P = 0.02; Analysis 1.14), mobility scale (MD 18.20, 95% CI 7.54 to 28.86; P = 0.0008; Analysis 1.17), and general recovery scale (MD 15.10, 95% CI 8.44 to 21.76; P < 0.00001; Analysis 1.23).

For SIS hand function at follow‐up (Barzel 2015), we found no significant effect (MD ‐2.20, 95% CI ‐12.46 to 8.06; P = 0.67; Analysis 2.14). These findings were based on one trial with a small number of participants resulting in a very low quality of evidence. The reported effects on SIS hand function were based on two trials with low risk of bias, but with clinical heterogeneity between studies, resulting in a moderate quality of evidence.

Measures of fatigue of the patient

None of the trials reported on effects of CME on fatigue of the patient after intervention or at follow‐up.

Length of stay

None of the included trials reported length of stay as an outcome measure. However, Galvin 2011 did state that mean length of hospital stay for the intervention group was 35.7 days (SD 10.5) and for the control group was 40.1 days (SD 15). Mean length of stay in a rehabilitation unit was 40.3 days (SD 9.6) for the intervention group and 52.3 days (SD 40) for the control group. Patients were recruited in a hospital and a rehabilitation unit. We found no significant differences for length of stay in a hospital (MD 4.40 days, 95% CI ‐3.91 to 12.71; P = 0.30; Analysis 1.24) or length of stay in a rehabilitation unit (MD 12.0 days, 95% CI ‐10.88 to 34.88; P = 0.30; Analysis 1.25). These effects were based on one trial, and length of stay was reported for a small number of participants (n = 20). Therefore, we graded the quality of the evidence as very low.

Adverse outcomes

One trial reported falls among participants (Dai 2013). We found no significant effect of CME on the number of falls reported (MD 0.04, 95% CI ‐0.10 to 0.18; P = 0.57; Analysis 1.26). There was no follow‐up in this trial. This effect was based on one trial with unclear risk of bias and a small number of participants, resulting in a very low quality of evidence.

Caregiver: measures of mood and quality of life

None of the included trials reported measures of mood or QoL of the caregiver.

Other outcomes

See Table 3.

Table 3. Results 'other outcomes' (not included in meta‐analysis)

Outcome	Control group (mean (SD))			Intervention group (mean (SD))
Outcome	Baseline	Post intervention	Follow‐up	Baseline	Post intervention	Follow‐up
Behavioural Inattention Test Conventional (Dai 2013)	48.79 (44.64)	68.83 (44.72)	‐	49.71 (39.63)	88.71 (44.56)	‐
Motor Assessment Scale (Galvin 2011)	29.7 (12.9)	34.5 (11.6)	35.2 (10.8)	24.3 (11.1)	36.1 (10.2)	37.9 (9.7)

SD: standard deviation.

Wall 1987 reported on gait parameters such as duration of single support phase and asymmetry ratio. We did not summarise these findings because they were beyond the scope of this review.

Dose of training

In three trials, the dose of training was comparable between the intervention and control groups (Abu Tariah 2010; Souza 2015; Wall 1987).

In six trials, the dose of training in the intervention group was higher than the dose of training in the control group (Agrawal 2013; Barzel 2015; Dai 2013; Galvin 2011; Gómez 2014; Wang 2015). In four of these trials, there was as higher dose of training in the intervention group because the intervention was additional to usual care and the control group received only usual care (Agrawal 2013; Dai 2013; Galvin 2011; Gómez 2014). In one trial, the intensity of training in the intervention group was higher due to the differences between interventions in the intervention and control groups (Wang 2015). The study compared a 90‐minute visit of a therapist and performing activities at least twice weekly, and if possible, every day in the intervention group, with a weekly visit or telephone call of the therapist and maintaining daily routines in the control group. In one trial, daily CIMT, which is a high‐intensity training intervention, was compared with usual care (Barzel 2015). With that, the intensity of training in the intervention group was higher than the dose of training in the control group.

We could not perform subgroup analysis for dose of training (higher dose of training versus same dose of training). For most outcome measures, all included trials had a higher dose of training in the intervention group, so no comparison could be made. For walking speed and upper arm function (Wolf Motor Function test and Motor Activity Log), one included trial was in the higher dose of training group and one included trial was in the same dose of training group. Because there was only one study per subgroup for these outcome measures, we did not perform a subgroup analysis.

Timing post stroke (Comparison 3)

We performed subgroup analyses for trials that included patients within six months after stroke (Agrawal 2013; Dai 2013; Galvin 2011; Gómez 2014) versus trials that included patients beyond six months after stroke (Barzel 2015; Wall 1987; Wang 2015). One trial included patients from beyond two months after stroke (Abu Tariah 2010), and another included patients directly after stroke (Souza 2015); however, the reported mean time since stroke was about nine months after stroke in the Abu Tariah 2010 study and 30 months after stroke in the Souza 2015 study. Therefore, we included both trials in the chronic phase group.

Because of the low number of included trials, we could only perform a subgroup analysis for the outcome measure basic ADL at the end of intervention. We found no difference between trials that included participants within six months after stroke when compared with trials that included patients beyond six months after stroke (P = 0.21; Analysis 3.1). The quality of evidence for this comparison was low, due to clinical heterogeneity between studies and a small total number of participants per subgroup.

For all other outcome measures, the number of included trials per subgroup was too low to test for subgroup differences.

Upper and lower extremity

Five trials were aimed at the upper extremity (Abu Tariah 2010; Agrawal 2013; Barzel 2015; Gómez 2014; Souza 2015), and four of these trials were about CIMT (Abu Tariah 2010; Barzel 2015; Gómez 2014; Souza 2015). However, Agrawal 2013, Gómez 2014, and Souza 2015 were not included in meta‐analysis.

Two trials were specifically aimed at the lower extremity (Galvin 2011; Wall 1987).

Basic and extended ADL were the only outcome measures in common when comparing upper and lower extremity trials. Due to the low number of trials per subgroup, we could not perform a subgroup analysis.

Reported mean changes (Comparison 4)

Mean change from post intervention to follow‐up

Galvin 2011 reported mean change at follow‐up (three months after termination of the intervention) from post intervention, using the outcome measures BI, CSI, NEADL Index, Reintegration to Normal Living Index, FMA lower extremity score, BBS, Six‐Minute Walk Test, and the Motor Assessment Scale.

This study found a significant effect in favour of CME for the Reintegration to Normal Living Index, CSI and the Six‐Minute Walk Test. The other mean changes were not significantly different. This result was based on one trial with a small number of participants, resulting in a very low quality of evidence.

Sensitivity analysis (Comparisons 5 and 6)

CME‐core

In five trials, CME was the only intervention (CME‐core) (Agrawal 2013; Galvin 2011; Souza 2015; Wall 1987; Wang 2015). Four trials studied the effect of another, existing intervention provided by the caregiver (Abu Tariah 2010; Barzel 2015; Dai 2013; Gómez 2014). In these four trials, it was difficult to separate the effects of CME from the effects of the other intervention (e.g. CIMT). Therefore, we performed a sensitivity analysis that included only CME‐core trials

Three CME‐core trials were suitable for meta‐analyses (Galvin 2011; Wall 1987; Wang 2015). We found a significant summary effect for basic ADL post intervention in favour of CME (2 studies; MD 9.45, 95% CI 2.11 to 16.78; P = 0.01; Analysis 5.1). This effect was based on two studies with low risk of bias, but with a small total number of participants for this outcome measure, resulting in a moderate quality of evidence.

We found no significant effect at follow‐up (1 study; MD 9.00, 95% CI ‐1.29 to 19.29; P = 0.09; Analysis 6.1). This effect was based on one study with a small number of participants, resulting in a very low quality of evidence.

For extended ADL, we found no significant summary effect post intervention (1 study; MD 5.50, 95% CI ‐5.83 to 16.83; P = 0.34; Analysis 5.2) or at follow‐up (1 study; MD 9.50, 95% CI ‐1.83 to 20.83; P = 0.10; Analysis 6.2). These effects were based on one study with a small number of participants resulting in a very low quality of evidence. For outcome measures relating to caregiver burden, we found no significant differences between the CME and control groups (see Analysis 1.3; moderate quality of evidence; and Analysis 2.3: very low quality of evidence).

For the secondary outcome measures, we found significant effects in favour of CME post intervention for standing balance (2 studies; MD 6.35, 95% CI 1.64 to 11.06; P = 0.008; Analysis 5.3; moderate quality of evidence) and QoL, concerning the composite physical subscale (1 study; MD 12.40, 95% CI 1.67 to 23.13; P = 0.02; Analysis 1.14; very low quality of evidence), mobility subscale (1 study; MD 18.20, 95% CI 7.54 to 28.86; P = 0.0008; Analysis 1.17; very low quality of evidence), and general recovery subscale of the SIS (1 study; MD 15.10, 95% CI 8.44 to 21.76; P < 0.00001; Analysis 1.23; very low quality of evidence). We found a significant effect in favour of CME for walking distance at follow‐up (1 study; MD 109.50 m, 95% CI 17.12 to 201.88; P = 0.02; Analysis 2.7; very low quality of evidence).

The included trials did not report the outcome measures FMA upper extremity, upper limb activities or function, length of stay, and adverse outcome for this sensitivity analysis.

The total number of included trials per subgroup within this sensitivity analysis was too small to test for subgroup differences.

Robustness of the results

In all analyses where we applied a fixed‐effect model, we subsequently applied a random‐effects model. This did not affect the overall results.

For one study, we combined two intervention groups (CME; CME plus physiotherapy) and two control groups (physiotherapy; no intervention) (Wall 1987). When we made separate comparisons of each intervention group versus each control group, we found no differences in the results (Comparison 7).

One study reported two outcome measures for extended ADL: the NEADL Index and the Reintegration to Normal Living Index (Galvin 2011). To prevent double counting this trial in our meta‐analysis, we included the NEADL Index in our primary analysis. We performed a sensitivity analysis in which we replaced the NEADL Index with the Reintegration to Normal Living Index (Comparison 8). Changing the outcome measure did not affect the direction or magnitude of the effect, neither did it affect the significance level of the meta‐analysis.

Qualitative synthesis

We could not include three trials in meta‐analyses: the various reasons are described in the Results of the search section (Agrawal 2013; Gómez 2014; Souza 2015).

All three trials were aimed at the upper extremity, with two trials applying CIMT (Gómez 2014; Souza 2015). For details of these trials, see the Characteristics of included studies table.

Agrawal 2013 comprised exercise training for the upper extremity in addition to usual care for two months. The three groups included a total of 30 participants. The results of each group are separately summarised in Table 4.

Table 4. Results Agrawal 2013 (study not included in meta‐analysis)

Outcome	Control group (mean scores)		GRASP 60 group (mean scores)		GRASP 90 group (mean scores)
Outcome	Baseline	Post intervention	Baseline	Post intervention	Baseline	Post intervention
Fugl‐Meyer Assessment upper extremity	31.3	37.0	32.9	44.0	34.7	48.2
Chedoke Arm and Hand Activities Inventory	20.3	26.8	21.0	30.0	24.4	37.0

Gómez 2014 studied CIMT with a caregiver in addition to usual care compared with usual care alone. The trial included a total of 60 participants and the intervention lasted 14 days. The goal of this trial was to determine if family support could increase eligibility for CIMT and to study the influence of social and family support. Reported outcomes were a description of the included participants and their level of social and family support. Furthermore, correlations were calculated between ADL, cognitive functioning, and level of social and family support, and were all found to be significant. Means and SDs were not reported. Gómez 2014 concluded that family can play a crucial role in delivering a CIMT protocol and that social and family support has a positive influence on functional outcome of the patient.

Souza 2015 studied CIMT (partly) performed together with a caregiver versus CIMT performed with a therapist. The trial included a total of 24 participants and had a follow‐up of six months. The study authors published effectiveness indexes for the outcome measures Motor Activity Log ‐ quality of movement, FMA upper extremity scale, and Stroke Specific Quality of Life Scale (SSQoL). There were no differences between experimental and control groups and the authors concluded that CIMT therapy (partly) together with a caregiver is equally effective as CIMT therapy with a therapist, but less expensive.

Discusión

available in

Resumen de los resultados principales

Para una visión general de los resultados, ver Resumen de los hallazgos para la comparación principal.

Efectos sobre las medidas de resultado

Esta revisión tuvo como objetivo determinar la efectividad de los EMC versus control en pacientes con accidente cerebrovascular. Se incluyeron nueve de 46 ensayos potencialmente relevantes. Los metanálisis incluyeron 333 parejas paciente‐cuidador. Cuatro ensayos evaluaron la medida de resultado primaria AC. No se encontró un efecto global significativo sobre las AC básicas al final de la intervención (Análisis 1.1; pruebas de calidad moderada) o al seguimiento (Análisis 2.1; pruebas de baja calidad). Para las AC ampliadas hubo dos ensayos en los que no se encontró un efecto global significativo al final de la intervención (Análisis 1.2; pruebas de baja calidad)o seguimiento (Análisis 2.2; pruebas de baja calidad). Dos ensayos evaluaron la medida de resultado primaria carga para el cuidador al final de la intervención y un ensayo al seguimiento. En ambos puntos temporales no se encontró un efecto global significativo de los EMC (al final de la intervención: Análisis 1.3; pruebas de calidad moderada; al seguimiento: Análisis 2.3; pruebas de muy baja calidad).

Con respecto a las medidas de resultado secundarias se encontró un efecto significativo a favor de los EMC al final de la intervención sobre la estabilidad postural (tres estudios; Análisis 1.6; pruebas de baja calidad) y la CdV (un estudio: subescalas física compuesta (Análisis 1.14), de movilidad (Análisis 1.17) y de recuperación general (Análisis 1.23); pruebas de muy baja calidad). La escala física compuesta es una puntuación de la suma de las escalas de fuerza, función manual, movilidad y AC/ACI. Se encontró un efecto significativo sobre la distancia de caminata al seguimiento (un estudio; Análisis 2.7; pruebas de muy baja calidad). En la prueba Wolf Motor Function ‐ tiempo de respuesta al final de la intervención hubo un efecto significativo a favor del grupo control (dos estudios; Análisis 1.10; pruebas de baja calidad). No se encontraron efectos significativos en la distancia de caminata después de la intervención ni en la estabilidad postural al seguimiento, y la CdV no se informó al seguimiento. No se encontraron efectos significativos en las puntuaciones de la FMA de la extremidad superior e inferior, la velocidad de marcha, las medidas de actividad o función del miembro superior, la duración de la estancia hospitalaria ni los eventos adversos (caídas) después de la intervención y el seguimiento (cuando se evaluaron). Ninguno de los ensayos incluidos informó sobre medidas de fatiga del paciente ni del estado de ánimo y la CdV del cuidador.

Desafortunadamente, debido al escaso número de ensayos incluidos no fue posible realizar análisis de subgrupos con respecto a la dosis de entrenamiento y al objetivo del entrenamiento con EMC dirigidos a la extremidad superior o inferior. En el análisis de subgrupos relacionado con el tiempo transcurrido desde el accidente cerebrovascular (en el transcurso de seis meses después del accidente cerebrovascular versus más de seis meses después del accidente cerebrovascular), sólo fue posible hacer una comparación en las AC básicas al final de la intervención. En las otras medidas de resultado el número de estudios incluidos por subgrupo fue demasiado pequeño. El tiempo transcurrido desde el accidente cerebrovascular no tuvo un efecto sobre las AC básicas al final de la intervención (Análisis 3.1; pruebas de baja calidad).

Un ensayo informó cambios medios desde después de la intervención hasta el seguimiento. La mayoría de los cambios medios informados favorecieron a los EMC. El cambio medio de la carga para el cuidador después de la intervención hasta el seguimiento estuvo significativamente a favor del grupo de EMC (Análisis 4.4; pruebas de muy baja calidad).

Estudios centrados en los EMC

En el análisis primario se incluyeron todos los ensayos de EMC. Sin embargo, varios ensayos proporcionaron los EMC como intervención única (centrados en los EMC), mientras que en otros el cuidador proporcionó una intervención existente, por ejemplo la TMIR. En estos últimos ensayos es difícil separar los efectos de los EMC de los efectos de la otra intervención.

El análisis de sensibilidad con los tres estudios centrados en los EMC mostró una diferencia importante en comparación con el análisis primario. Se encontró un efecto significativo a favor de los estudios centrados en los EMC en las AC básicas después de la intervención (Análisis 5.1; pruebas de calidad moderada). En las medidas de resultado secundarias se encontraron los mismos efectos significativos a favor de los EMC, al igual que en el análisis primario al final de la intervención para la estabilidad postural (Análisis 5.3; pruebas de calidad moderada) y la CdV (escala física compuesta: Análisis 1.14; movilidad: Análisis 1.17; escala de recuperación general: Análisis 1.23; todas con pruebas de calidad muy baja) y al seguimiento para la distancia de caminata (Análisis 2.7; pruebas de muy baja calidad). No fue posible realizar un análisis de subgrupos.

Es importante señalar que en el análisis de los ensayos centrados en los EMC sólo fue posible incluir los ensayos de la extremidad inferior. Un resultado de las AC, como el BI, es más sensible a la mejoría de la extremidad inferior que a la mejoría de la extremidad superior (Kwakkel 2004).

Estos efectos positivos de los ensayos centrados en los EMC sobre las AC básicas y la estabilidad postural pueden indicar una independencia mejor y más temprana, similar a las intervenciones de alta hospitalaria temprana con apoyo.

En este metanálisis se incluyó un número limitado de ensayos y medidas de resultado. Debido al número de participantes del ensayo de Barzel (n = 156), este ensayo tiene un efecto grande sobre los resultados (Barzel 2015). Las principales medidas de resultado afectadas son las AC básicas, las AC ampliadas y las medidas de actividades o función del miembro superior (prueba Wolf Motor Function y Motor Activity Log).

En este ensayo, la TMIR proporcionada por un cuidador se comparó con el tratamiento estándar; por lo tanto, este es uno de los ensayos en los que los efectos de los EMC son difíciles de separar de los efectos de la otra intervención (TMIR). Por lo tanto, se considera que el análisis de sensibilidad en el que solamente se incluyeron los ensayos centrados en los EMC es especialmente importante. Los efectos encontrados en el análisis de los ensayos centrados en los EMC son probablemente los más consistentes para responder el objetivo de esta revisión.

Importancia del análisis de los ensayos centrados en los EMC

Actividades cotidianas

No se encontraron efectos significativos sobre las AC básicas ni ampliadas en los análisis primarios. Estos resultados no fueron consistentes porque los EMC tuvieron un efecto positivo significativo sobre las AC básicas al final de la intervención en el análisis de sensibilidad de los ensayos centrados en los EMC. No hubo un efecto positivo sobre las AC básicas al seguimiento. Lo anterior se puede atribuir al efecto de techo de las medidas de resultado de las AC básicas (Quinn 2011). Por lo tanto, es importante que las medidas de las AC ampliadas se incluyan en los estudios que investigan los EMC.

Los EMC tiene la posibilidad de aumentar la intensidad del entrenamiento. En la mayoría de los ensayos incluidos en esta revisión los EMC aumentaron la intensidad del entrenamiento (Agrawal 2013; Barzel 2015; Dai 2013; Galvin 2011; Gómez 2014; Wang 2015). Varias revisiones sistemáticas han indicado que una intensidad mayor de entrenamiento puede dar lugar a un mejor resultado en los pacientes con accidente cerebrovascular en cuanto a las AC (French 2010; Galvin 2008a; Kwakkel 2004; Kwakkel 2006; Langhorne 2011; Lohse 2014; Veerbeek 2011; Veerbeek 2014), y, por lo tanto, se pueden esperar resultados favorables en cuanto a las AC. Sin embargo, a partir del escaso número de ensayos de prueba de concepto y las pruebas de calidad moderada, los resultados aún no son concluyentes y se necesitan más ensayos que evalúen las AC.

Carga para el cuidador

Los EMC son otra carga para el cuidador y, por lo tanto, se podría formular la hipótesis de que provocarán un aumento en la para el cuidador. Sin embargo, varios autores han alegado que en realidad la carga para el cuidador podría disminuir durante los EMC, debido a la educación concurrente del paciente y del cuidador y al mayor apoyo a los cuidadores al proporcionarle a los cuidadores más conocimiento acerca de las capacidades del paciente con accidente cerebrovascular y de ellos mismos (Galvin 2011; Kalra 2004; Wang 2015). Este hecho puede aumentar potencialmente los sentimientos de autoeficacia y control del cuidador (van den Heuvel 2001). Cuando en esta revisión se combinaron los datos de dos ensayos que evaluaron la carga para el cuidador no se encontraron efectos significativos, o sea, no hubo aumento ni disminución en la carga para el cuidador (Galvin 2011; Wang 2015). La calidad de las pruebas fue moderada. El cambio medio informado en el CSI después de la intervención y hasta el seguimiento estuvo a favor de los EMC (Galvin 2011).

En el metanálisis de los ensayos acerca de los EMC, Veerbeek 2014 mostró un tamaño del efecto positivo homogéneo y significativo sobre el esfuerzo del cuidador. La diferencia con el análisis de la presente revisión es que incluyeron a Kalra 2004, que se excluyó de esta revisión porque se aplicó una definición diferente de EMC. Kalra 2004 aplicó un entrenamiento de habilidades del cuidador, que en sentido estricto no es lo mismo que EMC ya que no es una intervención de entrenamiento progresivo. Sin embargo, como el entrenamiento de habilidades y los EMC pueden estar estrechamente relacionados, los resultados actuales sobre el efecto de los EMC sobre la carga para el cuidador no son consistentes. Por lo tanto, los resultados en la carga para el cuidador no son concluyentes y se necesitan más ensayos que evalúen este resultado en los EMC.

Cumplimiento de la seguridad

El cumplimiento de la seguridad es fundamental en los EMC. Solamente un ensayo incluido evaluó los eventos adversos en cuanto al número de caídas y no hubo diferencias entre los grupos de intervención y control. Estos resultados indican que, como mínimo, los EMC son tan seguros como la atención habitual. Sin embargo, la calidad de las pruebas fue muy baja. Debido a que el cuidador no es un terapeuta profesional, se necesitan métodos de cribado, entrenamiento e instrucción específicos para abordar los riesgos de seguridad (p.ej. las caídas). Por lo tanto, una parte importante de cada protocolo de EMC debe analizar la seguridad durante los EMC.

Dosis de entrenamiento

Veerbeek 2014 encontró pruebas sólidas a favor de las intervenciones de fisioterapia con entrenamiento intensivo, altamente repetitivo, orientado a tareas y específico a tareas en todas las fases después del accidente cerebrovascular. Lo anterior coincide con varios otros metanálisis que mostraron que la intensidad del entrenamiento y el entrenamiento en tareas repetitivas son los aspectos fundamentales de la rehabilitación del accidente cerebrovascular, lo que indica que más terapia con ejercicios es mejor (French 2010; Galvin 2008a; Kwakkel 2004; Kwakkel 2006; Langhorne 2011; Lohse 2014; Pollock 2014a; Veerbeek 2011; Veerbeek 2014). Pollock y colegas indicaron que una dosis de 30 a 60 minutos por día, proporcionada cinco a siete días por semana, tiene un valor adicional en cuanto a las actividades. Sin embargo, no es posible establecer conclusiones con respecto a la duración total de la intervención debido a la heterogeneidad significativa en los análisis (Pollock 2014a).

Todos los ensayos incluidos en el grupo de entrenamiento de alta intensidad tuvieron una dosis de al menos 30 a 60 minutos por día, administrado de cinco a siete días por semana. En todos los ensayos excepto Dai 2013, el grupo de intervención recibió al menos 16 horas de terapia con ejercicios en comparación con el grupo control (Kwakkel 2004; Veerbeek 2011). En un ensayo, el grupo de intervención recibió diez horas de tratamiento adicionales en comparación con el grupo control (Dai 2013).

Desafortunadamente, en la presente revisión no fue posible realizar un análisis de subgrupos del aumento de la dosis de entrenamiento en comparación con los ensayos pareados por dosis.

Tiempo transcurrido desde el accidente cerebrovascular

Los dos primeros meses después del accidente cerebrovascular se consideran el momento óptimo para la recuperación de la función (Cramer 2008; Hankey 2007; Jørgensen 1995; Jørgensen 1999; Kwakkel 2003; van Kordelaar 2014). Pollock 2014a encontró pruebas de un mayor efecto beneficioso de la intervención asociado con un tiempo más corto desde el accidente cerebrovascular. Por lo tanto, el aumento de la intensidad del entrenamiento con EMC parece especialmente significativo en los primeros meses después del accidente cerebrovascular. Sólo fue posible realizar un análisis de subgrupos (AC básicas después de la intervención), y no se encontraron diferencias entre los participantes que comenzaron la intervención en los seis primeros meses después del accidente cerebrovascular y los que comenzaron la intervención más de seis meses después del accidente cerebrovascular. Sin embargo, no es posible concluir si hay diferencias en el efecto de los EMC en diferentes puntos temporales después del accidente cerebrovascular debido al escaso número de ensayos incluidos en los análisis de subgrupos.

Extremidad superior versus inferior

Cinco de los nueve ensayos incluidos se dirigieron a la extremidad superior (Abu Tariah 2010; Agrawal 2013; Barzel 2015; Gómez 2014; Souza 2015), y cuatro estuvieron relacionados con la TMIR (Abu Tariah 2010; Barzel 2015; Gómez 2014; Souza 2015). La TMIR ha probado ser una terapia efectiva (Nijland 2011). Sin embargo, la TMIR puede ser una terapia que lleve mucho tiempo, y requerir la ayuda de un cuidador puede disminuir el tiempo con el terapeuta, por lo que la intervención aún es aplicable. Souza y colegas realizaron un ensayo importante que comparó la TMIR (en parte) junto con un cuidador con la TMIR proporcionada por un terapeuta. No encontraron diferencias entre los grupos experimental y control y concluyeron que estas formas de provisión de terapia son igualmente efectivas, pero que el entrenamiento con un cuidador es menos costoso en comparación con el entrenamiento con un terapeuta (Souza 2015).

En el análisis primario de la presente revisión se encontró un efecto significativo a favor de la intervención control sobre el tiempo de respuesta de la prueba Wolf Motor Function después de la intervención, pero no al seguimiento. Este resultado estuvo determinado en gran parte por un único estudio con un gran número de participantes (Barzel 2015), por lo que se debería considerar con cuidado.

Sólo dos ensayos incluidos se dirigieron específicamente a la extremidad inferior (Galvin 2011; Wall 1987). Una desventaja de las intervenciones dirigidas a la extremidad inferior es el aspecto de la seguridad. El riesgo de eventos adversos (p.ej. tropiezos o caídas) es mucho mayor cuando se practica el ponerse de pie o caminar en comparación con la práctica del uso de la extremidad superior. Sin embargo, las pruebas de los ensayos de intensidad centrados en el miembro inferior mostraron que fueron más efectivos que los que se dirigieron al miembro superior parético después del accidente cerebrovascular. Este último resultado hace que sea de interés centrar los EMC en la marcha o en las actividades relacionadas con la marcha.

Compleción y aplicabilidad general de las pruebas

Se encontró un número limitado de ensayos (nueve) con variaciones significativas en el tipo de EMC, la duración, el tiempo de entrenamiento (es decir, en el transcurso de seis meses o más de seis meses después de accidente cerebrovascular) y las medidas de resultado, lo que dificultó el resumen y la combinación de los datos en un metanálisis. Sin embargo, para ambas medidas de resultado primarias se encontraron dos o más ensayos de relativamente buena calidad.

Debido al número limitado de estudios incluidos, no hubo ensayos de calidad suficientemente buena disponibles para realizar los análisis de subgrupos, excepto para el tiempo transcurrido desde el accidente cerebrovascular (es decir, en el transcurso de seis meses o más de seis meses) para el resultado AC básicas.

Dos estudios incluyeron cuidadores retribuidos y no retribuidos, que no fue posible separar en los resultados (Dai 2013; Wang 2015). Por lo tanto, en esta revisión no fue posible comparar el efecto de los cuidadores retribuidos y no retribuidos. Los efectos de realizar ejercicios con un cuidador retribuido pueden ser diferentes en comparación con un cuidador no retribuido, especialmente cuando hay diferencias en la relación entre el paciente y el cuidador.

Puede haber diferencias culturales, étnicas y sociales entre las regiones y los países que pueden influir en la aplicabilidad y la efectividad de las intervenciones con EMC. Cuando el grupo étnico en sí puede no ser una limitación para los programas de EMC individualizados después del accidente cerebrovascular, pueden existir posibles facilitadores y barreras que se relacionan con la capacidad del profesional para sortear de manera efectiva las diferencias culturales y étnicas (Norris 2014).

Además, la participación de los cuidadores durante el proceso de rehabilitación puede ser más o menos fácil de implementar y puede ser más o menos aceptada como algo obvio en ciertas culturas por varias razones. En algunos países los servicios de rehabilitación no están fácilmente disponibles y se necesita la ayuda comunitaria, y a estos servicios que se les llama "rehabilitación en la comunidad" (WHO CBR). Uno de los ensayos excluidos realizó los EMC en ambos grupos (Redzuan 2012). Cuando se estableció contacto, el autor del estudio explicó que en Malasia es frecuente que se les pida ayuda a los cuidadores (o a trabajadores retribuidos). Por el contrario, los cuidadores en las culturas occidentales, con sistemas de asistencia sanitaria avanzados y diferentes prácticas sociales pueden ser más propensos a dejarles los servicios de asistencia sanitaria a los profesionales. Sin embargo, debido a los cambios constantes y a los recortes presupuestarios en la asistencia sanitaria occidental, existe más presión para que la familia brinde la atención. Por lo tanto, los EMC podrían tener implicaciones muy diferentes en diferentes culturas.

Calidad de la evidencia

El riesgo de sesgo de los nueve ensayos incluidos fue generalmente bajo o incierto (tabla "Características de los estudios incluidos). Desafortunadamente, no hubo datos suficientes (menos de diez ensayos) para examinar de manera confiable los efectos del riesgo de sesgo sobre las estimaciones del efecto mediante gráficos en embudo.

La calidad general de la evidencia fue de muy baja a moderada. Los detalles de los niveles GRADE de las pruebas se presentan en la sección "Efectos de las intervenciones". El metanálisis sólo pudo incluir seis ensayos que fueron pequeños, si se considera el número de participantes incluidos por ensayo. Por lo tanto, en la mayoría de las pruebas la calidad se disminuyó un nivel debido al número total pequeño de participantes (menos de 200 participantes). En algunas medidas de resultado (dirigidas principalmente al funcionamiento de la extremidad superior), hubo heterogeneidad estadística significativa no explicada y la calidad de las pruebas se disminuyó un nivel. En otras medidas de resultado hubo heterogeneidad clínica significativa. Hay una variación importante entre los ensayos en cuanto al tipo de ejercicio realizado con un cuidador. Se diferenció entre los ensayos de TMIR (Abu Tariah 2010; Barzel 2015), los ensayos con ejercicios de movilidad (Galvin 2011; Wall 1987; Wang 2015), y otros ensayos (Dai 2013). Cuando se combinaron estos ensayos, el nivel de las pruebas se disminuyó porque hubo heterogeneidad clínica. Además, puede haber sesgo de publicación en las comparaciones acerca del funcionamiento de la extremidad superior porque todos los ensayos no incluidos en los metanálisis estaban dirigidos al funcionamiento de la extremidad superior, por lo que el nivel de las pruebas se disminuyó en estas medidas de resultado.

Ver resumen de la calidad de las pruebas por medida de resultado en "Resumen de los hallazgos para la comparación principal".

Sesgos potenciales en el proceso de revisión

En algunos países los EMC parecen ser más necesarios o son más aceptados, o ambos, en la práctica diaria debido a la falta de servicios de rehabilitación formales o debido a actitudes culturales. Aunque es especulativo, la implementación de los EMC podría, por lo tanto, ser diferente en los países, lo que indica que en los ensayos de EMC el cumplimiento se debe medir sistemáticamente. Como no se identificaron ensayos completados de, por ejemplo, África, Asia y América del Sur, aún no ha sido posible encontrar información sobre dichas diferencias culturales.

En la revisión actual se diferenció entre los EMC y el entrenamiento de habilidades del cuidador, por lo que se excluyeron los ensayos sobre el entrenamiento de habilidades porque no se relaciona específicamente con una pareja que realiza ejercicios juntos. Podría haber potencialmente cierta superposición entre estas dos formas de entrenamiento. Al excluir los ensayos sobre entrenamiento de habilidades puede haberse perdido información potencialmente útil de estos ensayos. Sin embargo, existe confianza en que los resultados actuales reflejan de manera adecuada los efectos de los EMC.

Con respecto al análisis de los datos se utilizó la imputación o los procedimientos de extrapolación cuando no se informaron las DE o no fue posible obtenerlas de los autores de los estudios. En cuatro análisis se utilizaron las DE del mismo ensayo, por ejemplo, al inicio. Para la media del cambio se utilizó el IC del 95% y la puntuación Z para calcular las DE. Aunque lo anterior podría ser una posible fuente de sesgo, es poco probable que tuviera una repercusión importante sobre los resultados.

Figure 1

Study flow diagram.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Analysis 1.1

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 1 Patient: activities of daily living (ADL) measures: combined.

Analysis 1.2

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 2 Patient: ADL measures: extended ADL: combined.

Analysis 1.3

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 3 Caregiver: burden: combined.

Analysis 1.4

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 4 Measures of motor impairment: Fugl‐Meyer Assessment lower extremity.

Analysis 1.5

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 5 Measures of motor impairment: Fugl‐Meyer Assessment upper extremity.

Analysis 1.6

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 6 Gait and gait‐related measures: balance: combined.

Analysis 1.7

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 7 Gait and gait‐related measures: Six‐Minute Walk Test.

Analysis 1.8

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 8 Gait and gait‐related measures: walking speed.

Analysis 1.9

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 9 Measures of upper limb activities or function: Wolf Motor Function test ‐ functional ability.

Analysis 1.10

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 10 Measures of upper limb activities or function: Wolf Motor Function Test ‐ performance time.

Analysis 1.11

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 11 Measures of upper limb activities or function: Motor Activity Log (MAL) ‐ amount of use.

Analysis 1.12

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 12 Measures of upper limb activities or function: MAL ‐ quality of movement.

Analysis 1.13

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 13 Measures of upper limb activities or function: Nine Hole Peg test.

Analysis 1.14

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 14 Measures of mood and quality of life (QoL) of the patient: Stroke Impact Scale (SIS) ‐ composite physical.

Analysis 1.15

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 15 Measures of mood and QoL of the patient: SIS ‐ strength.

Analysis 1.16

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 16 Measures of mood and QoL of the patient: SIS ‐ ADL/IADL.

Analysis 1.17

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 17 Measures of mood and QoL of the patient: SIS ‐ mobility.

Analysis 1.18

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 18 Measures of mood and QoL of the patient: SIS ‐ hand function.

Analysis 1.19

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 19 Measures of mood and QoL of the patient: SIS ‐ memory.

Analysis 1.20

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 20 Measures of mood and QoL of the patient: SIS ‐ communication.

Analysis 1.21

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 21 Measures of mood and QoL of the patient: SIS ‐ emotion.

Analysis 1.22

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 22 Measures of mood and QoL of the patient: SIS ‐ social participation.

Analysis 1.23

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 23 Measures of mood and QoL of the patient: SIS ‐ general recovery.

Analysis 1.24

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 24 Length of stay ‐ hospital.

Analysis 1.25

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 25 Length of stay ‐ rehabilitation unit.

Analysis 1.26

Comparison 1 Caregiver‐mediated exercises versus control ‐ end of intervention, Outcome 26 Adverse outcomes: falls.

Analysis 2.1

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 1 Patient: activities of daily living (ADL) measures: ADL.

Analysis 2.2

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 2 Patient: ADL measures: extended ADL: combined.

Analysis 2.3

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 3 Caregiver: burden.

Analysis 2.4

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 4 Measures of motor impairment: Fugl‐Meyer Assessment lower extremity.

Analysis 2.5

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 5 Measures of motor impairment: Fugl‐Meyer Assessment upper extremity.

Analysis 2.6

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 6 Gait and gait‐related measures: balance.

Analysis 2.7

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 7 Gait and gait‐related measures: Six‐Minute Walking Test.

Analysis 2.8

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 8 Gait and gait‐related measures: walking speed.

Analysis 2.9

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 9 Measures of upper limb activities or function: Wolf Motor Function test ‐ functional ability.

Analysis 2.10

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 10 Measures of upper limb activities or function: Wolf Motor Function test ‐ performance time.

Analysis 2.11

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 11 Measures of upper limb activities or function: Motor Activity Log ‐ amount of use.

Analysis 2.12

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 12 Measures of upper limb activities or function: Motor Activity Log ‐ quality of movement.

Analysis 2.13

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 13 Measures of upper limb activities or function: Nine Hole Peg test.

Analysis 2.14

Comparison 2 Caregiver‐mediated exercises versus control ‐ end of follow‐up, Outcome 14 Measures of mood and quality of life of the patient: Stroke Impact Scale (SIS) ‐ hand function.

Analysis 3.1

Comparison 3 Timing post stroke ‐ end of intervention, Outcome 1 Patient: activities of daily living measures: combined.

Analysis 4.1

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 1 Patient: activities of daily living (ADL) measures: Barthel Index.

Analysis 4.2

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 2 Patient: ADL measures: extended ADL ‐ Nottingham Extended Activities of Daily Living Index.

Analysis 4.3

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 3 Patient: ADL measures: extended ADL ‐ reintegration to normal living index.

Analysis 4.4

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 4 Caregiver: Caregiver Strain Index.

Analysis 4.5

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 5 Measures of motor impairment: Fugl‐Meyer Assessment lower extremity.

Analysis 4.6

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 6 Gait and gait‐related measures: balance: Berg Balance Scale.

Analysis 4.7

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 7 Gait and gait‐related measures: Six‐Minute Walking Test.

Analysis 4.8

Comparison 4 Mean change from post intervention ‐ end of follow‐up, Outcome 8 Other outcomes: Motor Assessment Scale.

Analysis 5.1

Comparison 5 Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of intervention, Outcome 1 Patient: activities of daily living (ADL) measures: Barthel Index.

Analysis 5.2

Comparison 5 Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of intervention, Outcome 2 Patient: ADL measures: extended ADL ‐ Nottingham Extended Activities of Daily Living Index.

Analysis 5.3

Comparison 5 Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of intervention, Outcome 3 Gait and gait‐related measures: balance: Berg Balance Scale.

Analysis 6.1

Comparison 6 Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of follow‐up, Outcome 1 Patient: activities of daily living (ADL) measures: Barthel Index.

Analysis 6.2

Comparison 6 Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of follow‐up, Outcome 2 Patient: ADL measures: extended ADL ‐ Nottingham Extended Activities of Daily Living Index.

Analysis 7.1

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 1 Walking speed ‐ caregiver‐mediated exercises (CME) vs physiotherapy ‐ end of intervention.

Analysis 7.2

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 2 Walking speed ‐ CME vs physiotherapy ‐ end of follow‐up.

Analysis 7.3

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 3 Walking speed ‐ CME vs no intervention ‐ end of intervention.

Analysis 7.4

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 4 Walking speed ‐ CME vs no intervention ‐ end of follow‐up.

Analysis 7.5

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 5 Walking speed ‐ CME and physiotherapy vs physiotherapy ‐ end of intervention.

Analysis 7.6

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 6 Walking speed ‐ CME and physiotherapy vs physiotherapy ‐ end of follow‐up.

Analysis 7.7

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 7 Walking speed ‐ CME and physiotherapy vs no intervention ‐ end of intervention.

Analysis 7.8

Comparison 7 Walking speed, different possibilities study of Wall, Outcome 8 Walking speed ‐ CME and physiotherapy vs no intervention ‐ end of follow‐up.

Analysis 8.1

Comparison 8 Extended activities of daily living (ADL) ‐ analyses with Reintegration to Normal Living Index (RNLI), Outcome 1 Patient: ADL measures: extended ADL ‐ combined ‐ end of intervention.

Analysis 8.2

Comparison 8 Extended activities of daily living (ADL) ‐ analyses with Reintegration to Normal Living Index (RNLI), Outcome 2 Patient: ADL measures: extended ADL ‐ combined ‐ end of follow‐up.

Summary of findings for the main comparison. Caregiver‐mediated exercises compared with control intervention for people with stroke

Caregiver‐mediated exercises compared with control intervention for people with stroke
Patient or population: people with stroke Settings: inpatient and outpatient settings Intervention: caregiver‐mediated exercises Comparison: control, i.e. usual care, other intervention, no intervention
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control intervention	Caregiver‐mediated intervention
Patient: ADL measures Barthel Index. Scale 0 to 100 (follow‐up: 2 studies; 3/6 months) FIM. Scale 7 to 126 (no follow‐up)	The mean Barthel Index score ranged across control groups from 78 to 84 1 study: The mean FIM score in the control group was 65	The mean Barthel Index score in the intervention groups was 5.09 higher (‐2.88 to 13.07 higher) 1 study: The mean FIM score in the intervention group was 11 higher (‐1.59 to 23.67 higher)	‐	Barthel Index: 247 (3) FIM: 48 (1) Total: 295	⊕⊕⊕⊝ Moderate	Higher scores are better More than half of the studies at low risk of bias (3 low risk of bias, 1 at unclear risk of bias) There was clinical heterogeneity SMD 0.21 (‐0.02 to 0.44)
Caregiver: measures of mood, burden and QoL: burden Caregiver Strain Index Scale. 0 to 13 (follow‐up 3 months) Caregiver Burden Scale. 22 to 88 (no follow‐up)	The mean Caregiver Strain Index score in the control group was 3.4 The mean Caregiver Burden Scale score in the control group was 46.6	The mean Caregiver Strain Index score in the intervention group was 0.50 higher (‐0.81 to 1.81 higher) The mean Caregiver Burden Scale score in the intervention group was 1.30 lower (‐4.88 to 7.48 lower)	‐	Caregiver Strain Index: 40 (1) Caregiver Burden Scale: 51 (1) Total: 91	⊕⊕⊕⊝ Moderate	Lower scores are better Both studies at low risk of bias Small total number of participants SMD ‐0.04 (‐0.45 to 0.37)
Gait and gait‐related measures: walking speed in m/s (follow‐up: 1 study, 9 months)	The mean walking speed ranged across control groups from 0.26 m/s to 0.46 m/s	The mean walking speed in the intervention group was 0.08 m/s higher (‐0.03 to 0.18)	‐	71 (2)	⊕⊝⊝⊝ Very low
Gait and gait‐related measures: walking distance measured with the Six‐Minute Walk Test in metres walked in 6 minutes (follow‐up: 1 study, 3 months)	The mean distance walked ranged across control groups from 157 m to 166 m	The mean distance walked in the intervention groups was 30.98 m higher (‐20.22 to 82.19 higher)	‐	91 (2)	⊕⊕⊕⊝ moderate	Lower scores are better 1 study at unclear risk of bias Small total number of participants MD 0.04 (‐0.10 to 0.18)
Measures of mood and QoL of the patient: Stroke Impact Scale Stroke Impact Scale mobility scale. Scale 9 to 45. (no follow‐up)	The mean Stroke Impact Scale mobility score in the control group was 66.8	The mean Stroke Impact Scale mobility score in the intervention group was 18.2 higher (7.54 to 28.86 higher)	‐	51 (1)	⊕⊝⊝⊝ Very low	Higher scores are better 1 study at low risk of bias Small total number of participants MD 18.2 (7.54 to 28.86)
Length of stay: length of stay in rehabilitation unit in days	The mean length of stay in a rehabilitation unit in the control group was 52.3 days	The mean length of stay in a rehabilitation unit in the intervention group was 12 days lower (‐10.88 to 34.88)	‐	20 (1)	⊕⊝⊝⊝ very low	Higher scores are better 1 study at low risk of bias and 1 at unclear or high risk of bias Small total number of participants There was clinical heterogeneity MD 0.08 m/s (‐0.03 to 0.18)
Adverse outcomes: falls number of falls/patient (no follow‐up)	1 study: the mean number of falls/patient in the control group was 0.08	1 study: the mean number of falls/ patient in the intervention group was 0.04 lower (‐0.10 to 0.18 lower)	‐	48 (1)	⊕⊝⊝⊝ Very low	Higher scores are better Both studies at low risk of bias Small total number of participants MD 30.98 m (‐20.22 to 82.19)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). ADL: activities of daily living; CI: confidence interval; FIM: Functional Independence Measure; MD: mean difference; QoL: quality of life; RR: risk ratio; SD: standard deviation; SMD: standardised mean difference.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Summary of findings for the main comparison. Caregiver‐mediated exercises compared with control intervention for people with stroke

Table 1. Outline of included studies

Study ID	Form of training	Upper or lower body	Timing since stroke	Task caregiver	Routine care continued	Control group	Programme (length ‐frequency‐ duration)	Contact with therapist	Place
Abu Tariah 2010	CIMT	Upper	> 2 months	Carried out the intervention with support of therapists	No	Neurodevelopmental training, same intensity	2 months ‐ daily ‐ 2 hours	3 or 4 sessions	Home
Agrawal 2013	Exercise therapy	Upper	"Sub‐acute stroke"	Encouragement, participating, and help	Yes	Usual care	4 weeks ‐5 days/week ‐ 60 to 90 minutes	Weekly	Inpatient?
Barzel 2015	CIMT	Upper	> 6 months	Supervision, help, and maintaining training diary	No	Usual care, frequency of seeing a therapist was the same	4 weeks ‐ Every weekday (not weekend) ‐ 2 hours	5 x 60 minutes	Home
Dai 2013	Vestibular rehabilitation	Both	< 6 months	Guidance and supervision (in third and fourth week)	Yes	Usual care	4 weeks ‐ 10 sessions per 2 weeks ‐ 30 minutes	2 to 4 sessions in first 2 weeks	Inpatient?
Galvin 2011	Exercise therapy	Lower	Assessment 2 weeks after stroke onset	Encouragement and help	Yes	Usual care	8 weeks ‐every day ‐ 35 minutes	Weekly	Inpatient or at home
Gómez 2014	CIMT	Upper	< 6 months	Monitoring and supervising	Yes	Usual care	14 days ‐ every day* ‐ 5.5 hours*	1.5 hours per day*	Inpatient
Souza 2015	CIMT: 1.5 hours with therapist and 1.5 hours with caregiver	Upper	< 24 months**	Supervision and making notes	No	CIMT: 3 hours with therapist	22 days ‐ 10 sessions ‐ 3 hours	10 x 90 minutes	Outpatient and home
Wall 1987	Exercise therapy	Lower	After discharge of rehabilitation	Supervision	No	No intervention	6 months ‐ twice a week ‐ 1 hour	1 group: twice a week 1 group: once a week 1 group: 'monitoring'	Outpatient or at home
Wang 2015	Exercise programme aimed at body functions, activities, and participation	Both	> 6 months	Encouragement and help	No	Usual care	12 weeks ‐ minimal twice a week, if possible every day ‐ minimal 50 to 60 minutes	Weekly 90 minutes	Home
CIMT: constraint‐induced movement therapy. * Details of the intervention are not completely clear, contact with the authors was not successful. ** But mean time since stroke was 27 and 35 months since stroke, unclear why.

Table 1. Outline of included studies

Table 2. (Standard) Mean differences which are not reported in section 'data and analysis'

Outcome	Outcome measure	Fixed‐effect or random‐effects model	Mean difference	Confidence interval	Heterogeneity	P value
1.1 Patient: ADL measures ‐ Combined	1.1.1 Barthel Index	Random‐effects	5.09	‐2.88 to 13.07	58%	0.21
1.1 Patient: ADL measures ‐ Combined	1.1.2 Functional Independence Measure	Fixed‐effect	11.04	‐1.59 to 23.67	‐	0.09
1.2 Patient: ADL measures ‐ extended ADL	1.2.1 Nottingham Extended Activities of Daily Living Index	Fixed‐effect	5.50	‐5.83 to 16.83	‐	0.34
1.2 Patient: ADL measures ‐ extended ADL	1.2.2 IADL	Fixed‐effect	0.02	‐0.72 to 0.76	‐	0.96
1.3 Caregiver: burden	1.3.1 Caregiver Strain Index	Fixed‐effect	‐0.50	‐1.81 to 0.81	‐	0.46
1.3 Caregiver: burden	1.3.2 Caregiver Burden Scale	Fixed‐effect	1.30	‐4.88 to 7.48	‐	0.68
1.6 Gait and gait‐related measures: balance	1.6.1 Berg Balance Scale	Fixed‐effect	6.35	1.64 to 11.06	0%	0.008
1.6 Gait and gait‐related measures: balance	1.6.2 Postural Assessment for Stroke patients	Fixed‐effect	3.50	‐0.52 to 7.52	‐	0.09
2.2 Patient: ADL measures ‐ extended ADL	2.2.1 Nottingham Extended Activities of Daily Living Index	Fixed‐effect	9.50	‐1.83 to 20.83	‐	0.10
2.2 Patient: ADL measures ‐ extended ADL	2.2.2 IADL	Fixed‐effect	0.02	‐0.77 to 0.81	‐	0.96
3.1 Patient: ADL measures ‐ combined	3.1.1 < 6 months	Fixed‐effect	0.44*	0.01 to 0.86	0%	0.04
3.1 Patient: ADL measures ‐ combined	3.1.2 > 6 months	Random‐effects	4.90	‐7.56 to 17.36	77%	0.44
8.1 Patient ADL measures ‐ extended ADL ‐ end of intervention	8.1.1 Reintegration to normal living Index	Fixed‐effect	0.20	‐3.76 to 4.16	‐	0.92
	8.1.2 IADL	Fixed‐effect	0.02	‐0.72 to 0.76	‐	0.96
8.2 Patient ADL measures ‐ extended ADL ‐ end of follow‐up	8.2.1 Reintegration to normal living Index	Fixed‐effect	4.50	0.54 to 8.46	‐	0.03
8.2 Patient ADL measures ‐ extended ADL ‐ end of follow‐up	8.2.2 IADL	Fixed‐effect	0.02	‐0.77 to 0.81	‐	0.96
ADL: activities of daily living; IADL: instrumental activities of daily living. *Standardised mean difference.

Table 2. (Standard) Mean differences which are not reported in section 'data and analysis'

Table 3. Results 'other outcomes' (not included in meta‐analysis)

Outcome	Control group (mean (SD))			Intervention group (mean (SD))
Outcome	Baseline	Post intervention	Follow‐up	Baseline	Post intervention	Follow‐up
Behavioural Inattention Test Conventional (Dai 2013)	48.79 (44.64)	68.83 (44.72)	‐	49.71 (39.63)	88.71 (44.56)	‐
Motor Assessment Scale (Galvin 2011)	29.7 (12.9)	34.5 (11.6)	35.2 (10.8)	24.3 (11.1)	36.1 (10.2)	37.9 (9.7)
SD: standard deviation.

Table 3. Results 'other outcomes' (not included in meta‐analysis)

Table 4. Results Agrawal 2013 (study not included in meta‐analysis)

Outcome	Control group (mean scores)		GRASP 60 group (mean scores)		GRASP 90 group (mean scores)
Outcome	Baseline	Post intervention	Baseline	Post intervention	Baseline	Post intervention
Fugl‐Meyer Assessment upper extremity	31.3	37.0	32.9	44.0	34.7	48.2
Chedoke Arm and Hand Activities Inventory	20.3	26.8	21.0	30.0	24.4	37.0

Table 4. Results Agrawal 2013 (study not included in meta‐analysis)

Comparison 1. Caregiver‐mediated exercises versus control ‐ end of intervention

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: activities of daily living (ADL) measures: combined Show forest plot	4	295	Std. Mean Difference (IV, Fixed, 95% CI)	0.21 [‐0.02, 0.44]

1.1 Barthel Index	3	247	Std. Mean Difference (IV, Fixed, 95% CI)	0.16 [‐0.09, 0.41]
1.2 Functional Independence Measure	1	48	Std. Mean Difference (IV, Fixed, 95% CI)	0.49 [‐0.09, 1.06]
2 Patient: ADL measures: extended ADL: combined Show forest plot	2	196	Std. Mean Difference (IV, Fixed, 95% CI)	0.07 [‐0.21, 0.35]

2.1 Nottingham Extended Activities of Daily Living Index	1	40	Std. Mean Difference (IV, Fixed, 95% CI)	0.29 [‐0.33, 0.92]
2.2 Instrumental Activities of Daily Living (IADL)	1	156	Std. Mean Difference (IV, Fixed, 95% CI)	0.01 [‐0.31, 0.32]
3 Caregiver: burden: combined Show forest plot	2	91	Std. Mean Difference (IV, Fixed, 95% CI)	‐0.04 [‐0.45, 0.37]

3.1 Caregiver Strain Index	1	40	Std. Mean Difference (IV, Fixed, 95% CI)	‐0.23 [‐0.85, 0.39]
3.2 Caregiver Burden Scale	1	51	Std. Mean Difference (IV, Fixed, 95% CI)	0.11 [‐0.44, 0.66]
4 Measures of motor impairment: Fugl‐Meyer Assessment lower extremity Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	3.10 [‐2.02, 8.22]

5 Measures of motor impairment: Fugl‐Meyer Assessment upper extremity Show forest plot	1	18	Mean Difference (IV, Fixed, 95% CI)	4.43 [‐2.09, 10.95]

6 Gait and gait‐related measures: balance: combined Show forest plot	3	139	Std. Mean Difference (IV, Fixed, 95% CI)	0.53 [0.19, 0.87]

6.1 Berg Balance Scale	2	91	Std. Mean Difference (IV, Fixed, 95% CI)	0.56 [0.14, 0.98]
6.2 Postural Assessment Scale for Stroke Patients	1	48	Std. Mean Difference (IV, Fixed, 95% CI)	0.48 [‐0.09, 1.06]
7 Gait and gait‐related measures: Six‐Minute Walk Test Show forest plot	2	91	Mean Difference (IV, Fixed, 95% CI)	30.98 [‐20.22, 82.19]

8 Gait and gait‐related measures: walking speed Show forest plot	2	71	Mean Difference (IV, Fixed, 95% CI)	0.08 [‐0.03, 0.18]

9 Measures of upper limb activities or function: Wolf Motor Function test ‐ functional ability Show forest plot	2	174	Mean Difference (IV, Random, 95% CI)	0.02 [‐0.52, 0.55]

10 Measures of upper limb activities or function: Wolf Motor Function Test ‐ performance time Show forest plot	2	174	Mean Difference (IV, Fixed, 95% CI)	‐1.72 [‐2.23, ‐1.21]

11 Measures of upper limb activities or function: Motor Activity Log (MAL) ‐ amount of use Show forest plot	2	174	Mean Difference (IV, Fixed, 95% CI)	0.01 [‐0.36, 0.38]

12 Measures of upper limb activities or function: MAL ‐ quality of movement Show forest plot	2	174	Mean Difference (IV, Fixed, 95% CI)	0.08 [‐0.26, 0.42]

13 Measures of upper limb activities or function: Nine Hole Peg test Show forest plot	1	156	Mean Difference (IV, Fixed, 95% CI)	‐0.04 [‐0.11, 0.03]

14 Measures of mood and quality of life (QoL) of the patient: Stroke Impact Scale (SIS) ‐ composite physical Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	12.40 [1.67, 23.13]

15 Measures of mood and QoL of the patient: SIS ‐ strength Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	12.20 [‐0.08, 24.48]

16 Measures of mood and QoL of the patient: SIS ‐ ADL/IADL Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	11.4 [‐1.11, 23.91]

17 Measures of mood and QoL of the patient: SIS ‐ mobility Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	18.20 [7.54, 28.86]

18 Measures of mood and QoL of the patient: SIS ‐ hand function Show forest plot	2	207	Mean Difference (IV, Fixed, 95% CI)	2.64 [‐5.87, 11.15]

19 Measures of mood and QoL of the patient: SIS ‐ memory Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	6.30 [‐1.65, 14.25]

20 Measures of mood and QoL of the patient: SIS ‐ communication Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	3.0 [‐2.34, 8.34]

21 Measures of mood and QoL of the patient: SIS ‐ emotion Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	2.10 [‐4.35, 8.55]

22 Measures of mood and QoL of the patient: SIS ‐ social participation Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	6.70 [‐1.69, 15.09]

23 Measures of mood and QoL of the patient: SIS ‐ general recovery Show forest plot	1	51	Mean Difference (IV, Fixed, 95% CI)	15.10 [8.44, 21.76]

24 Length of stay ‐ hospital Show forest plot	1	37	Mean Difference (IV, Fixed, 95% CI)	4.40 [‐3.91, 12.71]

25 Length of stay ‐ rehabilitation unit Show forest plot	1	20	Mean Difference (IV, Fixed, 95% CI)	12.0 [‐10.88, 34.88]

26 Adverse outcomes: falls Show forest plot	1	48	Mean Difference (IV, Fixed, 95% CI)	0.04 [‐0.10, 0.18]

Comparison 1. Caregiver‐mediated exercises versus control ‐ end of intervention

Comparison 2. Caregiver‐mediated exercises versus control ‐ end of follow‐up

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: activities of daily living (ADL) measures: ADL Show forest plot	2	196	Mean Difference (IV, Random, 95% CI)	2.69 [‐8.18, 13.55]

1.1 Barthel Index	2	196	Mean Difference (IV, Random, 95% CI)	2.69 [‐8.18, 13.55]
2 Patient: ADL measures: extended ADL: combined Show forest plot	2	196	Std. Mean Difference (IV, Fixed, 95% CI)	0.11 [‐0.17, 0.39]

2.1 Nottingham Extended Activities of Daily Living Index	1	40	Std. Mean Difference (IV, Fixed, 95% CI)	0.51 [‐0.12, 1.14]
2.2 IADL	1	156	Std. Mean Difference (IV, Fixed, 95% CI)	0.01 [‐0.31, 0.32]
3 Caregiver: burden Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	0.60 [‐0.71, 1.91]

3.1 Caregiver Strain Index	1	40	Mean Difference (IV, Fixed, 95% CI)	0.60 [‐0.71, 1.91]
4 Measures of motor impairment: Fugl‐Meyer Assessment lower extremity Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	3.40 [‐1.74, 8.54]

5 Measures of motor impairment: Fugl‐Meyer Assessment upper extremity Show forest plot	1	18	Mean Difference (IV, Fixed, 95% CI)	2.75 [‐8.24, 13.74]

6 Gait and gait‐related measures: balance Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	8.40 [‐1.04, 17.84]

6.1 Berg Balance Scale	1	40	Mean Difference (IV, Fixed, 95% CI)	8.40 [‐1.04, 17.84]
7 Gait and gait‐related measures: Six‐Minute Walking Test Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	109.50 [17.12, 201.88]

8 Gait and gait‐related measures: walking speed Show forest plot	1	20	Mean Difference (IV, Fixed, 95% CI)	0.10 [‐0.02, 0.22]

9 Measures of upper limb activities or function: Wolf Motor Function test ‐ functional ability Show forest plot	2	174	Mean Difference (IV, Random, 95% CI)	0.08 [‐0.46, 0.61]

10 Measures of upper limb activities or function: Wolf Motor Function test ‐ performance time Show forest plot	2	174	Mean Difference (IV, Random, 95% CI)	1.85 [‐8.78, 12.48]

11 Measures of upper limb activities or function: Motor Activity Log ‐ amount of use Show forest plot	2	174	Mean Difference (IV, Random, 95% CI)	0.21 [‐0.65, 1.08]

12 Measures of upper limb activities or function: Motor Activity Log ‐ quality of movement Show forest plot	2	174	Mean Difference (IV, Fixed, 95% CI)	‐0.03 [‐0.43, 0.37]

13 Measures of upper limb activities or function: Nine Hole Peg test Show forest plot	1	156	Mean Difference (IV, Fixed, 95% CI)	‐0.05 [‐0.12, 0.02]

14 Measures of mood and quality of life of the patient: Stroke Impact Scale (SIS) ‐ hand function Show forest plot	1	156	Mean Difference (IV, Fixed, 95% CI)	‐2.20 [‐12.46, 8.06]

Comparison 2. Caregiver‐mediated exercises versus control ‐ end of follow‐up

Comparison 3. Timing post stroke ‐ end of intervention

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: activities of daily living measures: combined Show forest plot	4	295	Std. Mean Difference (IV, Fixed, 95% CI)	0.21 [‐0.02, 0.44]

1.1 < 6 months	2	88	Std. Mean Difference (IV, Fixed, 95% CI)	0.44 [0.01, 0.86]
1.2 > 6 months	2	207	Std. Mean Difference (IV, Fixed, 95% CI)	0.12 [‐0.16, 0.39]

Comparison 3. Timing post stroke ‐ end of intervention

Comparison 4. Mean change from post intervention ‐ end of follow‐up

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: activities of daily living (ADL) measures: Barthel Index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	2.3 [‐3.95, 8.55]

2 Patient: ADL measures: extended ADL ‐ Nottingham Extended Activities of Daily Living Index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	4.00 [‐0.99, 8.99]

3 Patient: ADL measures: extended ADL ‐ reintegration to normal living index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	4.3 [2.03, 6.57]

4 Caregiver: Caregiver Strain Index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	1.1 [0.45, 1.75]

5 Measures of motor impairment: Fugl‐Meyer Assessment lower extremity Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	0.30 [‐2.21, 2.81]

6 Gait and gait‐related measures: balance: Berg Balance Scale Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	‐0.9 [‐4.78, 2.98]

7 Gait and gait‐related measures: Six‐Minute Walking Test Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	43.3 [15.11, 71.49]

8 Other outcomes: Motor Assessment Scale Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	1.1 [‐0.92, 3.12]

Comparison 4. Mean change from post intervention ‐ end of follow‐up

Comparison 5. Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of intervention

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: activities of daily living (ADL) measures: Barthel Index Show forest plot	2	91	Mean Difference (IV, Fixed, 95% CI)	9.45 [2.11, 16.78]

2 Patient: ADL measures: extended ADL ‐ Nottingham Extended Activities of Daily Living Index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	5.5 [‐5.83, 16.83]

3 Gait and gait‐related measures: balance: Berg Balance Scale Show forest plot	2	91	Mean Difference (IV, Fixed, 95% CI)	6.35 [1.64, 11.06]

Comparison 5. Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of intervention

Comparison 6. Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of follow‐up

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: activities of daily living (ADL) measures: Barthel Index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	9.0 [‐1.29, 19.29]

2 Patient: ADL measures: extended ADL ‐ Nottingham Extended Activities of Daily Living Index Show forest plot	1	40	Mean Difference (IV, Fixed, 95% CI)	9.5 [‐1.83, 20.83]

Comparison 6. Sensitivity analysis ‐ caregiver‐mediated exercise (CME)‐core ‐ end of follow‐up

Comparison 7. Walking speed, different possibilities study of Wall

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Walking speed ‐ caregiver‐mediated exercises (CME) vs physiotherapy ‐ end of intervention Show forest plot	2	61	Mean Difference (IV, Fixed, 95% CI)	0.06 [‐0.07, 0.20]

2 Walking speed ‐ CME vs physiotherapy ‐ end of follow‐up Show forest plot	1	10	Mean Difference (IV, Fixed, 95% CI)	0.11 [‐0.04, 0.26]

3 Walking speed ‐ CME vs no intervention ‐ end of intervention Show forest plot	2	61	Mean Difference (IV, Fixed, 95% CI)	0.06 [‐0.08, 0.19]

4 Walking speed ‐ CME vs no intervention ‐ end of follow‐up Show forest plot	1	10	Mean Difference (IV, Fixed, 95% CI)	0.10 [‐0.04, 0.24]

5 Walking speed ‐ CME and physiotherapy vs physiotherapy ‐ end of intervention Show forest plot	2	61	Mean Difference (IV, Fixed, 95% CI)	0.07 [‐0.06, 0.21]

6 Walking speed ‐ CME and physiotherapy vs physiotherapy ‐ end of follow‐up Show forest plot	1	10	Mean Difference (IV, Fixed, 95% CI)	0.10 [‐0.11, 0.31]

7 Walking speed ‐ CME and physiotherapy vs no intervention ‐ end of intervention Show forest plot	2	61	Mean Difference (IV, Fixed, 95% CI)	0.07 [‐0.07, 0.20]

8 Walking speed ‐ CME and physiotherapy vs no intervention ‐ end of follow‐up Show forest plot	1	10	Mean Difference (IV, Fixed, 95% CI)	0.09 [‐0.11, 0.29]

Comparison 7. Walking speed, different possibilities study of Wall

Comparison 8. Extended activities of daily living (ADL) ‐ analyses with Reintegration to Normal Living Index (RNLI)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Patient: ADL measures: extended ADL ‐ combined ‐ end of intervention Show forest plot	2	196	Std. Mean Difference (IV, Fixed, 95% CI)	0.01 [‐0.27, 0.29]

1.1 RNLI	1	40	Std. Mean Difference (IV, Fixed, 95% CI)	0.03 [‐0.59, 0.65]
1.2 Instrumental Activities of Daily Living (IADL)	1	156	Std. Mean Difference (IV, Fixed, 95% CI)	0.01 [‐0.31, 0.32]
2 Patient: ADL measures: extended ADL ‐ combined ‐ end of follow‐up Show forest plot	2	196	Std. Mean Difference (IV, Random, 95% CI)	0.29 [‐0.37, 0.95]

2.1 RNLI	1	40	Std. Mean Difference (IV, Random, 95% CI)	0.69 [0.05, 1.33]
2.2 IADL	1	156	Std. Mean Difference (IV, Random, 95% CI)	0.01 [‐0.31, 0.32]

Comparison 8. Extended activities of daily living (ADL) ‐ analyses with Reintegration to Normal Living Index (RNLI)