Although we were unable to examine the impact of different modes of delivery of CIMT, such as signature CIMT (sCIMT), modified CMIT (mCIMT) or hybrid CMIT (hCIMT), our review shows that CIMT can be implemented in a range of modes and settings, that constraint can be achieved using various devices, and the accompanying intervention can be delivered by interventionists other than therapists, including parents and students. Our review indicates that the specific mode of CIMT intervention is a lesser issue than implementation of an intensive, carefully‐targeted and well‐supported programme. However, maintenance of treatment fidelity is essential. Clinicians should ensure that the two key ingredients across all models of CIMT are maintained: 1) restraint of the well‐functioning upper limb (irrespective of device/type); and 2) intensive, structured training (irrespective of type) (Eliasson 2014a). The mean number of hours of CIMT provided across studies was 129 hours (range = 20 hours in Yu 2012 to 504 hours in Sung 2005). We did not identify any study that concluded that short‐term constraint methods, such as occasional hand holding, were effective. No study provided CIMT for a period longer than 10 weeks. Clincians, therefore, should view CIMT as a relatively short‐term intervention that is provided for a defined period, and carefully evaluate outcomes before and after implementation using valid and reliable measures.

The high‐dose and dose‐matched comparison interventions, predominantly intensive occupational therapy and bimanual interventions, offer evidence‐based options for families. As with CIMT, these interventions should be implemented with adherence to intervention fidelity, including the nature of intervention and dose.

It is important for clinicians to educate children and families about the outcomes of this review. CIMT appears to be a safe intervention for children with unilateral CP. Families should feel confident that, on average, active engagement in a well‐defined, intensive program of CIMT or bimanual therapy can lead to improvements in bimanual performance and unimanual capacity. Discussions with families will include the magnitude of the benefit, the uncertainty of long‐term benefits of the blocks of intervention, and the need to continually monitor children’s upper‐limb function and occupational performance to identify appropriate timing of further episodes of intensive, upper‐limb interventions or implementation of alternative means of achieving child‐ and family‐centred goals. It should be emphasised that not all children respond to CIMT (Hoare 2015). The challenge remains for researchers to identify the most appropriate of these interventions to implement with individual children. This review was not able to identify the characteristics of children who could be advised to participate in one or the other of CIMT or bimanual interventions. In the meantime, clinicians should consider the specific goals for individual children and families and choose the most developmentally appropriate, family‐centred, and convenient of these approaches (Hoare 2017). Factors, in addition to child and family characteristics and preferences, which may impact of intervention selection include therapist expertise, costs of implementing the intervention, funding and service delivery models, and resource availability.

Inadequate reporting of both CIMT and comparison interventions was common and substantial in this review. We recommend that future trials use the Template for Intervention Description and Replication (TIDieR) checklist and guideline for reporting interventions (Hoffmann 2014). The TIDieR checklist guides study authors to provide details on the rationale for the interventions included in the study, materials required, procedures followed, and to specify who provided the intervention, modes of delivery, where, when and how much intervention was provided, any intervention tailoring available, modifications made to the intervention during study implementation, and planned and actual intervention adherence or fidelity. This information can be provided as supplementary information or included in a published trial protocol. A full description of all comparison interventions is required.

We located 86 published papers or abstracts (range 1 to 10 per study) reporting findings from the 36 studies included in this review. During the literature search and data extraction process, it was frequently difficult and time consuming to establish whether a publication resulted from an existing study and was a duplicate; reported a separate set of outcomes or contained a subgroup of participants; or was a unique study by the same group of authors. In the Agreements and disagreements with other studies or reviews section of this review, we have identified that other researchers have not recognised multiple publications from the same study and have inadvertently synthesised findings from duplicate publications of the same study. To avoid confusion, future studies of CIMT should consider publishing a single manuscript for reporting study results. At the very least, each publication emanating from the same study using the same cohort or subgroups of children should explicitly refer to previous publications and clearly articulate the relationship of the publication to the study as a whole. Trial registration is frequently mandatory and publication of a study protocol is becoming more common ‐ both will assist systematic reviewers and others synthesising evidence to identify and accurately analyse and interpret findings.

We recommend that all authors of trials investigating CIMT in children with CP follow the CONSORT guidelines for reporting RCTs (Schulz 2010) and the extended HARMS guidelines for reporting adverse events (Ioannidis 2004). Although relatively few adverse events potentially related to the CIMT were reported, only 50% of included studies reported monitoring this outcome and the absence of adverse events cannot be confirmed. Implementation of a standardised method of recording and reporting adverse events would ensure more consistent and deliberate reporting.

The choice of outcome measure should be carefully matched to the expected effect of the intervention in all research evaluating health‐related interventions. We strongly recommend future studies use reliable outcome measures that have been validated for children with CP and their families and are matched to the aims of CIMT ‐ improving unimanual capacity and bimanual performance. Understanding its impact on individualised goals related to self‐care and ability to complete other everyday activities is also relevant to consider. Fifty‐seven outcome measures were used by the 36 included studies included in this review and over half of these were used in a single study. This severely limits the ability to pool data for meta‐analysis, slowing the development of further knowledge in this area of research. Uniform follow‐up periods after completion of CIMT could also be adopted to enable more accurate meta‐analysis of studies. Studies should adhere to the standardised procedures for the administration and scoring of outcome measures. When these procedures are modified, the validity and reliability of the outcome is not maintained and the integrity of the measure is threatened (Eliasson 2014a). Unless studies are investigating the cumulative and longitudinal effects of multiple blocks of CIMT or bimanual therapy (or both), outcome measures should also reflect the potential impact of a short‐term intensive block of upper‐limb activity level intervention. Although CIMT has demonstrated domain‐specific changes in quality of life in some studies included in this review, a single block of CIMT does not aim to change multi‐dimensional constructs such as quality of life (Gilson 2014) or participation (Adair 2015; Imms 2016b). Selection of outcome measures in future studies of CIMT should reflect this and efforts should also be made to minimise the potential for assessment burden for children and their families. Research has also repeatedly demonstrated there is no evidence that CIMT either improves, or leads to deterioration in, body function and structure outcomes. Studies could justifiably avoid adding to assessment burden and research waste by refraining from further measurement of these types of variables.