Vagus nerve stimulation to improve post‐stroke motor function and activity

Objectives

This is a protocol for a Cochrane Review (intervention). The objectives are as follows:

To assess the effectiveness and safety of vagus nerve stimulation as an add‐on treatment to rehabilitate people with post‐stroke motor function impairments and activity limitations.

Background

Description of the condition

Stroke is defined as 'an episode of acute neurological dysfunction presumed to be caused by ischaemia or haemorrhage, persisting ≥ 24 h or until death' (Hankey 2017; Sacco 2013). Stroke is the second leading cause of death and third leading cause of disability (GBD 2021). People with hypertension, cardiac diseases, diabetes, hypercholesterolaemia, obesity; those who smoke, consume excessive alcohol or drugs; and those with a reduced level of physical activity face an increased risk of stroke (GBD 2021). According to recent predictions, one in four people will suffer a stroke during their lifetime (Gorelick 2019). Brain damage attributed to the loss of neurons caused by stroke can result in motor, sensory, and cognitive deficits. About half of the people with stroke present with upper limb motor impairment and cognitive disorders (Fekadu 2019). These disabilities may limit functional independence, reduce quality of life, and affect mental health (Gil‐Salcedo 2022).

Description of the intervention

Current Canadian guidelines recommend exercise therapy programmes to improve post‐stroke motor function (Teasell 2020). However, despite rehabilitation, half of the people will remain permanently impaired (Lee 2015). To promote recovery, interventions that enhance brain plasticity have been developed (Su 2020). Evidence has shown that the improvement of brain plasticity, characterised by increased neural activity and connectivity, is an excellent predictor of motor recovery (Stinear 2010). In this context, and to enhance outcomes of stroke rehabilitation, treatments involving vagus nerve stimulation (VNS) have recently been developed.

VNS involves any neuromodulation treatment that alters the activity of the tenth cranial nerve. VNS typically involves surgical implementation of a pulse generator, leads, and electrodes under the skin of the left pectoral and neck regions while the person is under general anaesthesia (Howland 2014). This subcutaneous system is coupled with an external programmable device that serves to set up the parameters of the electrical stimulation (Afra 2021). When set up, the electrical signal is transmitted from the pulse generator to the vagus nerve through the leads (Wheless 2018). VNS is generally regarded as safe and well tolerated. Adverse events, when they occur, are commonly associated with either the surgical procedure or the stimulation process itself. Adverse events related to stimulation include voice alteration, cough, dyspnoea, paraesthesia, headache, and local pain (Wheless 2018). With continued treatment, the frequency of these adverse events tends to diminish. Signal amplitude may vary from 0 to 3.5 mA, current frequency from 1 Hz to 30 Hz, and pulse width from 100 µs to 1000 µs (Wheless 2018). More recently, studies have described the use of non‐invasive VNS treatment, using a handheld stimulator over the skin covering the vagus nerve in the carotid triangle region (Arsava 2022), or in the auricular cavum conchae (Li 2022). Transcutaneous VNS is also considered safe and well tolerated. Skin irritation is the most prevalent adverse event (Redgrave 2018).

How the intervention might work

Overall, neurological recovery may be improved by enhancing the brain’s ability to reorganise post‐stroke residual networks (brain plasticity (Engineer 2019; Krakauer 2004; Nudo 2006)). While the precise mechanism of VNS remains unknown, it is thought to promote neuroplasticity by driving the rapid activation of neuromodulatory networks (cholinergic, serotoninergic, and noradrenergic systems), thanks to the nucleus tractus nerve solitarius projection of the vagus nerve (Ma 2019). Through this projection, the vagus nerve modulates the release of serotonin, acetylcholine, and norepinephrine from the dorsal raphe nucleus, nucleus basalis, and locus coeruleus to different brain regions (Gu 2002; Manta 2009). Recent evidence suggests that the predominant mechanism may be cholinergic reinforcement. The heightened noradrenergic activity induced by VNS is purported to influence the trisynaptic circuit in the hippocampus, a relay system involved in synaptic plasticity (Olsen 2023). VNS‐induced alterations in the locus coeruleus are closely tied to adrenergic signalling, and can potentiate perforant path‐dentate gyrus‐evoked potentials and enhance perforant path‐CA3 field excitatory post‐synaptic potentials (Shen 2012). This cascade of events is associated with long‐term potentiation, a form of synaptic plasticity linked to enhanced learning and memory (Kandel 2014). VNS boosts the excitability and spontaneous spiking in the CA1 region and enhances synaptic transmission (Olsen 2022). VNS also induces lasting changes in synaptic plasticity by influencing mediators, such as growth factors, immediate early genes, and proteins critical for long‐term potentiation (Olsen 2023). This includes an increase in the expression of brain‐derived neurotrophic factor (BDNF) and the activation of tropomyosin receptor kinase B, the receptor for BDNF (Furmaga 2012). Fibroblast growth factor expression is also augmented by VNS, along with enhanced expression of adrenergic receptors and proteins crucial for long‐term potentiation, such as GluN2B (NMDA receptor subunit) and its downstream signalling target calcium/calmodulin‐dependent protein kinase II (Olsen 2023). These comprehensive changes induced by VNS contribute to long‐term alterations in plasticity and memory consolidation in the hippocampus, highlighting the multifaceted mechanisms through which VNS may enhance cortical reorganisation. The activation of neuromodulatory networks through the vagus nerve is expected to have neuroprotective effects, as it limits oxidative stress and neuroinflammatory activation, and enhances cortical map reorganisation by strengthening synapse efficacy and triggering neurogenesis (Engineer 2019; Ma 2019).

Existing literature indicates that VNS enhances hippocampal synaptic plasticity in healthy individuals, and demonstrates improvements in attention and memory for people with epilepsy, when used as an antiepileptic intervention (Olsen 2023). But there is currently no documented evidence that supports the efficacy of VNS in influencing neuroplasticity in individuals with stroke. Nevertheless, animal studies have provided evidence supporting the positive effects of VNS on neuronal and corticospinal plasticity, therefore, underlining its potential to improve neurological recovery after a stroke (Ay 2009; Biggio 2009; Meyers 2018). These outcomes reinforce the perspective that VNS is contingent on neuroplasticity, a phenomenon that unfolds over time and is sensitive to temporal dynamics. Therefore, we may expect differences between the effect of VNS during the acute and chronic phases of people with stroke. The purpose of starting in the acute phase is to prevent the loss of motor function or the development of other disabilities, by protecting the neurons from breaking down. Recent findings strongly indicate that VNS could be an appealing approach to mitigate the adverse effects of spreading depolarisation waves in acute brain disorders (Cheng 2022; Lindemann 2020), therefore, underlining its potential to preserve the penumbra and restrict the extent of the infarct core in people with acute stroke (Dreier 2011; Hartings 2017; Nakamura 2010). During the chronic phase, the treatment goal is to promote neuroplasticity and induce neurogenesis to improve disabilities. A study conducted with rats with chronic stroke, showed that VNS did not diminish the size of the lesion, but improved the recovery process by enhancing neuroplasticity (Khodaparast 2016).

The positive effects of VNS on mood may also play a role in the context of stroke rehabilitation. Prior research indicated that mood symptoms following an acute stroke were correlated with a lower quality of life at one year (Donnellan 2010). Importantly, this study also revealed that depressive symptoms exerted a more pronounced influence on functional recovery. In this context, the mood‐enhancing benefits of VNS may not only contribute to an improved quality of life, but potentially, could also have a positive impact on functional recovery, underscoring the broader rehabilitative potential of VNS interventions in post‐stroke care.

When combined with motor training, animal studies reveal that VNS results in the reorganisation of cortical motor maps, affirming the capacity of VNS to foster synaptic plasticity when a targeted timing approach is used (Tseng 2020). In adults with stroke, the combination of VNS and high‐repetition, functional, and progressive exercise therapy showed promising results for upper limb motor function (Dawson 2021). A pivotal randomised sham‐controlled trial, involving 108 individuals with ischaemic stroke, showed that VNS paired with rehabilitation significantly improved upper limb motor control and activity. These changes were maintained for three months post‐intervention. The number of people achieving clinically significant improvement in upper limb control was greater in the VNS group than in the control group.

Why it is important to do this review

Potentially, VNS may be an adjunctive treatment for rehabilitation in people with stroke. Therefore, it remains important to assess the effectiveness and safety of VNS to improve post‐stroke motor function and activity, to aid future guidance for VNS implementation in stroke survivors. Six meta‐analyses have reviewed the effectiveness and safety of VNS in people after a stroke. None of them used a GRADE approach to analyse the certainty of the body of evidence (Ananda 2022; Liu 2022; Ramos‐Castaneda 2022; Xie 2021; Zhang 2020; Zhao 2021). Since the publication of the last meta‐analysis, two new randomised controlled trials, assessing the effect and safety of VNS in people with stroke, have been published (Arsava 2022; Li 2022). These two studies enroled a total of 129 participants, which represent about 62% of the total sample size of the last meta‐analysis. All participants were treated during the subacute phase (compared to 85% of participants in previous studies), and new outcomes were evaluated (magnetic resonance imaging outcomes, the National Institute of Health Stroke Scale score, and lower limb motor function). Li 2022 assessed participants with severe motor impairments and evaluated their outcomes during a one‐year follow‐up. The number of studies on non‐invasive VNS is also increasing.

Despite the findings of recent meta‐analyses, several factors remain unknown, which potentially influence the effectiveness of VNS, such as impairment severity and time since stroke onset. Some authors also questioned whether motor changes induced by VNS could be clinically important (Kwakkel 2021). If so, these therapies may also be used for other populations suffering from brain injury. To date, the distribution of VNS effects in the population and the implications for health equity remain underexplored. With the addition of new randomised controlled trials and the elaboration of subgroup analyses, it is worthwhile to synthesise the current evidence.

Objectives

To assess the effectiveness and safety of vagus nerve stimulation as an add‐on treatment to rehabilitate people with post‐stroke motor function impairments and activity limitations.

Methods

Criteria for considering studies for this review

Types of studies

We will include only randomised controlled trials (RCTs) as they are the strongest study design to evaluate the efficacy of interventions (McKenzie 2023). We will include parallel, cluster, and cross‐over trials. We will not include quasi‐RCTs.

Types of participants

We will include studies reporting on adults (age > 18 years old) who were diagnosed with their first stroke (ischaemic or hemorrhagic), as defined by the American Heart Association (Sacco 2013). We will include participants who are in hyperacute (0 to 24 hours since stroke onset), acute (1 day to 7 days), early subacute (7 days to 3 months), late‐subacute (3 months to 6 months), or chronic (> 6 months) phases (Bernhardt 2017).

We will exclude studies of participants with mixed pathologies (for example, studies that involved adults with a stroke and adults with dementia).

Types of interventions

Experimental interventions

We will include studies in which the experimental group received a vagus nerve stimulation (VNS) intervention, paired with usual care (as defined by the authors) or conventional rehabilitation (intervention during which a clinician provides and supervises the rehabilitation tasks; therapy recommended by the most recent guidelines).

In this review, we will consider a VNS intervention to be any technique involving the use of a device (with an invasive (placed under) or non‐invasive (over the skin) placement) to stimulate the vagus nerve with pulses of electrical energy. The signal amplitude of VNS usually varies from 0 to 3.5 mA, the current frequency from 1 Hz to 30 Hz, and the pulse width from 100 µs to 1000 µs. We will consider all VNS interventions, regardless of their signal amplitude, current frequency, or pulse width. There will be no restriction on treatment intensity, duration and number of sessions, or length and frequency of treatment, for any VNS. We will consider all types of coupling between VNS and motor training (VNS provided before, during, or after the trials).

Control interventions

We will include studies in which the control group received usual care (as defined by the authors), conventional rehabilitation, or another active intervention that aims to promote the motor recovery and activity of the participant’s affected limb. We will also include studies in which the control group received sham VNS or minimal VNS stimulation (maximum of one pulse greater than 0 mA per session). There will be no restriction on treatment intensity, duration and number of sessions, or length and frequency of treatment for the control interventions. We will consider all types of coupling between sham VNS and motor training (sham VNS provided before, during, or after the trials).

Types of outcome measures

We will select eligible studies that reported at least one of the following outcome measures.

Primary outcomes

Effectiveness of VNS

• Upper limb motor function: Upper Extremity Fugl‐Meyer Assessment (UE‐FMA), a scale ranging from 0 to 66 points, with a minimal clinically important difference (MCID) of 6.6 points (Fugl‐Meyer 1975)

Safety of VNS

• Number of participants with at least one serious adverse event (SAE); measured at the end of intervention and follow‐up; an SAE is any event that results in death, is life‐threatening, requires or prolongs hospitalisation, leads to significant or permanent disability, or causes congenital anomalies or birth defects (Cook 2008). We will consider events, such as infection, dysphagia, and vocal cord paresis as SAEs.

Timing of outcome assessment

We will consider the short‐term, medium‐term, and long‐term time points for the analysis of outcomes.

• Short‐term (post‐intervention outcome is closest to six weeks but does not exceed 12 weeks). We will use the short‐term time point as the primary time‐point of interest.

• Medium‐term (outcome closest to six months)

• Long‐term (outcome closest to 12 months)

Secondary outcomes

Effectiveness of VNS

For studies using more than one scale to measure a given outcome, we will use a hierarchical order of preference as recommended in the guidelines (Kwakkel 2017; Lanctôt 2020; Pohl 2020; Prange‐Lasonder 2021; Van Criekinge 2023; Winstein 2016). We present the prioritised scales first. The others are enclosed in parentheses and listed in order of preference.

• Upper limb activity: Action Research Arm Test (ARAT), a scale ranging from 0 to 57 points, with an MCID of 5.5 points (Yozbatiran 2008); (Wolf Motor Function Test, a scale ranging from 0 to 5 points, with an MCID of 0.4 points (Lin 2009; Wolf 2001))

• Spasticity: modified Ashworth Scale, a scale ranging from 0 to 4 points (Gregson 1999); (modified Tardieu Scale, a scale ranging from 0 to 4 points for each joint (Morris 2002); upper limb Brunnstorm stage, a scale ranging from 1 to 7 points (Brunnstrom 1966))

• Manual dexterity: Box and Block Test, a functional measure ranging from 0 to 150 points (Mathiowetz 1985a); (Nine‐Hole Peg Test, an average time measure ranging from 0 to 300 seconds (Mathiowetz 1985b))

• Lower limb motor function: Lower‐Extremity Fugl‐Meyer Assessment (LL‐FMA), a scale ranging from 0 to 34 points (Fugl‐Meyer 1975)

• Quality of life: Stroke Impact Scale (SIS), a scale ranging from 0% to 100% (Duncan 1999); (EuroQol 5‐dimension 5‐level, a multidimensional scale (Rabin 2001))

• Anxiety: Hospital Anxiety and Depression Scale (HADS), a scale ranging from 0 to 21 points (Zigmond 1983); (General Anxiety Disorder scale, a scale ranging from 0 to 21 points (Spitzer 2006))

• Stroke severity improvement: National Institute of Health Stroke Scale (NIHSS), a scale ranging from 0 to 42 (Brott 1989)

• Relative infarct growth: a magnetic resonance imaging (MRI) measure based on diffusion‐weighted imaging and computed as: (post‐intervention lesion volume‐baseline lesion volume/baseline lesion volume) × 100 (Arsava 2022; Dávalos 2004)

Safety of VNS

• Number of participants who discontinued the intervention (due to any reason related to the intervention)

• Number of participants with at least one AE; AEs include pain, fall, injury, etc.

Timing of outcome assessment

As for the primary outcomes, we will consider the short‐term, medium‐term, and long‐term time points for the analysis.

Search methods for identification of studies

Electronic searches

We will search for published trials in the following databases, using systematic search strategies.

• Cochrane Central Register of Controlled Trials (CENTRAL; current issue) in the Cochrane Library

• MEDLINE PubMed (1971 to current)

• Embase Elsevier (2007 to current)

• Scopus Elsevier (2004 to current)

• PsychINFO Ovid (1967 to current)

• CINAHL EBSCO (1977 to current)

• PEDro (1999 to current)

We developed a systematic search strategy for PubMed using MeSH terms and free terms. We then adapted this systematic search strategy for the other databases. All search strategies are presented in Appendix 1. To avoid publication bias, we will not restrict the searches by language or article type (Lefebvre 2023).

Searching other resources

We will search for ongoing and unpublished trials in opengrey.eu, clinicaltrial.gov, the World Health Organisation International Clinical Trials Registry Platform, and strokecenter.org/trials.

We will handsearch the reference lists of all included studies, relevant scoping and systematic reviews identified during the searches, and conference papers (Conference Proceedings Citation Index‐Science; 1990 to current) covering the use of VNS for stroke rehabilitation (Lefebvre 2023). We will also contact trial authors of the included studies to request information about ongoing or unpublished studies.

Review authors will also perform searches on Google Scholar, looking for studies that could have been missed, or that were not published in the above‐mentioned databases.

Data collection and analysis

Selection of studies

We will download all references from the searches and store them in Covidence systematic review software (Covidence; Lefebvre 2023). After removal of duplicates, two review authors (GE and IS) will independently perform the first selection of potentially relevant reports, based on titles and abstracts. This process will be followed by a meeting to resolve any conflicting selections between review authors. If no agreement is reached, another review author will make the final decision (GEB or BL, or both). The review authors (GE and IS) will independently review the full text of all remaining articles. They will meet again to resolve any conflicting selection, inviting a third review author (GEB or BL, or both) in case of discrepancy. During the full‐text review, we will record a brief list of excluded studies, which seemed to meet the eligibility criteria before full‐text review. For each of these studies, we will document the primary cause of exclusion (Lefebvre 2023). We will also search for potential errata statements of the included studies. Author retraction, fraud suspicion, or important limitations (calling into question the validity of the results) will lead to exclusion of the study (Li 2023).

Data extraction and management

Two reviewers (GE and IS) will independently extract the following data from each study, using a data collection form (Li 2023): number of participants randomised, number of dropouts, age, sex, time since stroke onset (or stroke stage classification), degree of severity of arm impairment (according to baseline score of UE‐FMA), design of the intervention (type of VNS treatment, VNS modalities, area targeted by VNS, coupling method, type of add‐on therapy, dose, duration, intensity, frequency), design of the control intervention (type of sham‐VNS treatment, sham‐VNS modalities, area targeted by sham‐VNS, coupling method, type of usual care, dose, duration, intensity, frequency), length of follow‐up assessment, primary and secondary outcome scores at baseline, post‐intervention, and follow‐up, main findings of the study, country in which the study was conducted, funding of the study, and possible conflicts of interest of authors. This form will be independently pilot tested by two review authors (GE and IS) using a representative sample of the studies to be reviewed.

For multi‐arm studies, we will only extract data from relevant intervention groups (Li 2023). Disagreements will be resolved through discussion. One review author (GE) will export all the data into the Review Manager file, and another review author (IS) will check that the data were exported correctly (RevMan 2024).

Assessment of risk of bias in included studies

Two review authors (GE and IS) will independently evaluate the risk of bias in each included study using RoB 2 (Boutron 2023; Sterne 2019). This tool enables the assessment of risk of bias for each outcome. We will evaluate the impact of assignment to the intervention for the short‐term outcomes of the following outcome measures: upper limb motor function (assessed with the UE‐FMA), upper limb activity (assessed with the ARAT), quality of life (assessed with the SIS), and SAE. The risk of bias assessment covers the following domains: bias arising from the randomisation process (domain 1a), bias due to deviations from intended interventions (domain 2), bias due to missing outcome data (domain 3), bias in measurement of the outcome (domain 4), and bias in selection of the reported result (domain 5). We will use specific variants of the RoB 2 to handle the risk of bias assessment for cross‐over and cluster‐RCTs (Higgins 2023a). These specific variants of RoB 2 include additional domains, such as bias arising from period effect and carryover effect (domain S) for cross‐over RCTs, and bias arising from the identification or recruitment of participants into cluster (domain 1b) for cluster‐RCTs.

For each domain, we will classify the overall risk of bias for each outcome as low, high, or with some concerns. In the case of uncertainty or disagreement between review authors, we will involve a third reviewer (GEB or BL, or both); if needed, we will contact the authors of the study.

We will use the Excel tool (available at riskofbiasinfo.org) to record and manage RoB 2 assessments. We will then produce risk of bias graphs and risk of bias summary graphs. We will provide support for the judgements and the sources of information for each risk of bias assessment (Boutron 2023).

Measures of treatment effect

We will report dichotomous outcomes, such as the number of dropouts, AEs, or SAEs using a risk ratio (RR) and 95% confidence interval (CI). For continuous outcomes, if the included studies used the same scale to measure the outcome, we will report the treatment effect using mean differences (MDs) and 95% CI. If different scales were used to measure the same outcome, we will measure treatment effectiveness by computing a standardised mean difference (SMD) and 95% CI (Higgins 2023b). We will compare post‐intervention outcomes at the end of treatment, and at follow‐up.

We will check the accuracy of numerical data by comparing the direction of effect between the RR, MD, or SMD, and the findings of the included studies (Li 2023).

Unit of analysis issues

If individuals underwent more than one intervention (due to a cross‐over design), we will only include the data that were reported before the cross‐over (first phase), since there is a risk of carryover due to the potential persistence of the effect of the first intervention (Higgins 2023a).

For three‐arm studies in which all intervention groups are relevant, we will give priority to the splitting approach (Rücker 2017; Welch 2023). This method divides the ‘shared’ group into subgroups of equal sample size. If the study compared two types of VNS interventions to a conventional therapy programme, the shared group would be the conventional therapy one. Conversely, if the study compared a VNS intervention to a conventional therapy program and no intervention, the shared group would be the VNS group. This method allows an effect comparison between all participants’ first group and half of the participants’ shared group; and between all participants’ second group and the other half of the participants’ shared group.

For cluster‐RCTs, we will determine the effective sample sizes for both the intervention and control groups. For this, we will compute a ratio between the original sample size and the design effect. We will approximate the design effect by adjusting the average cluster size through the intraclass correlation coefficient (ICC (Higgins 2023a)). If studies do not provide the ICC, we will impute this information based on comparable studies. Subsequently, we will conduct a sensitivity analysis to assess the impact of the imputation.

Dealing with missing data

In cases of missing data (either full‐text report unavailability or missing statistical information), we will contact the corresponding authors of the included studies. For continuous data, if trial authors are unable to provide the mean and standard deviations (SDs), we will convert the medians and their quartiles to means and SDs, using Wan’s approach (Wan 2014). Similarly, we will convert CIs and standard errors of the means to SDs following the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Boutron 2023).

Assessment of heterogeneity

We will quantify between‐study statistical heterogeneity using the I² statistic. We will interpret it as low (I² between 0% and 40%), moderate (I² between 40% and 60%), substantial (I² between 50% and 90%), or considerable (I² between 75% and 100% (Deeks 2023; Higgins 2002; Higgins 2023b)). We will also assess heterogeneity by visual inspection of forest plots, and with Chi² tests.

We will assess clinical heterogeneity using the characteristics of included studies tables to compare the population, interventions, comparisons, and outcomes. If the studies are sufficiently similar, we will pool the data. We anticipate subgroup analysis based on the stroke phase to investigate clinical heterogeneity. To this end, we will compare results between participants who are in the (hyper‐)acute‐early‐subacute phase (0 hours to 3 months post‐stroke onset) and participants who are in the late‐subacute‐chronic phase (> 3 months post‐stroke onset).

We will assess methodological heterogeneity using sensitivity analysis, to see whether including studies at a high risk of bias affects our conclusions.

Assessment of reporting biases

We will evaluate reporting biases by comparing the information described in each trial’s published protocol (in clinical trials registers) to the methods and results sections of the publication. If we find uncertainty or inconsistency between reports or sections, we will contact the study authors. We will also use a funnel plot to check for small study effects, which may suggest publication bias (through symmetry visual analyses), if we find more than 10 studies that reported the same outcome. If we find asymmetry or outliers (studies with markedly different intervention effect estimates), we will undertake the Egger regression test to assess the influence of publication bias on each result (Egger 1997).

Data synthesis

If, based on the clinical heterogeneity assessment, the studies are sufficiently similar, we will synthesise VNS treatment effectiveness for post‐stroke upper limb motor function, activity, spasticity, manual dexterity, quality of life, and anxiety; and safety (proportion of dropouts, AEs, and SAEs) using a meta‐analysis. This meta‐analysis will include all eligible studies. We will use RevMan software for all statistical analyses and forest plots. We will use a random‐effects model, as we expect some heterogeneity between studies. If we include fewer than two studies, or the heterogeneity is considerable (I² ≥ 75%), even after removing outliers (studies with markedly different intervention effect estimates), we will summarise the results using a narrative synthesis, according to Synthesis Without Meta‐analysis guidelines, instead of a pooled statistical synthesis (Campbell 2020).

Subgroup analysis and investigation of heterogeneity

If we have data from ≥ 10 studies, we will undertake subgroup analyses, using the formal test for subgroup differences in RevMan.

Because post‐lesional neuroplasticity is purported to be at its greatest within the first three months of brain injury, we will start with subgroup analyses to compare the findings collected in the (hyper‐)acute and early subacute phases (0 hours to 3 months post‐stroke onset) with those collected in the late subacute and chronic phases (> 3 months post‐stroke onset).

We will assess the influence of impairment severity (low‐moderate versus severe motor impairments) and VNS device set‐up (invasive versus non‐invasive) on post‐stroke motor and activity outcomes.

We will also assess health equity by determining whether the effect of VNS may change according to the location of studies. To this end, we will compare VNS effect on people with stroke from high‐, middle‐, and low‐income countries (Hawe 2004; Pope 2007).

Sensitivity analysis

We will conduct the following sensitivity analysis to assess the robustness of the decision we made in our analysis:

• Risk of bias: analyses limited to studies at low risk of bias and those with some concerns.

• Imputation: analyses excluding studies with imputed data

Summary of findings and assessment of the certainty of the evidence

We will develop a summary of findings table for the comparison:

• VNS as an add‐on therapy versus no training or usual care (with or without sham VNS) in people with stroke.

We will include these outcomes in the summary of findings table:

• Upper limb motor function

• Number of participants with at least one SAE

• Upper limb activity

• Quality of life

This summary of findings table will comprise the population and settings, the experimental and comparison intervention, the four listed health outcomes, the measure of each outcome (RR, MD, or SMD), the magnitude of effect, the numbers of participants and studies, and the overall certainty of the body of evidence for each outcome (Schünemann 2023).

We will report outcomes measured in the short‐term (≤ 3 months). However, if subgroup analyses reveal a significant difference between outcomes collected in the (hyper‐)acute‐early‐subacute phase, and those in the late‐subacute‐chronic phase, we will categorise outcomes separately in the summary of findings table, based on these phases.

Two review authors (GE and IS) will evaluate the certainty of the body of evidence for each outcome using the GRADE approach (GRADEpro GDT; Schünemann 2023). We will meet with one or more additional review authors (including GEB) to address any disagreements, ultimately reaching a consensus. GRADE rates the certainty of evidence for an outcome as high, moderate, low, or very low, according to: study limitations/overall risk of bias, inconsistency, indirectness, imprecision, and publication bias (Brożek 2009).

We will downgrade the certainty of the evidence by one level for serious and two levels for very serious concerns for each domain.

Appendices

Appendix 1. Search strategies

Cochrane Central Register of Controlled Trials in the Cochrane Library

Title Abstract Keyword: "Vagus Nerve Stimulation" OR "Vagus Nerve" OR "10th cranial nerve" OR "cranial nerve X" OR "non‐invasive vagus nerve" OR "transcutaneous auricular vagus nerve" OR "transcutaneous auricular nerve" OR "VNS" OR "ta‐VNS" OR "pneumo‐gastric nerve" OR "pneumo gastric nerve" OR "pneumogastric nerve" OR "stimulation of the vagus nerve" OR "tenth cranial nerve"

AND

Title Abstract Keyword: "stroke" OR "cva" OR "accident cerebrovascular" OR "acute cerebrovascular lesion" OR "brain vascular accident" OR "cerebral stroke" OR "cerebro vascular accident" OR "cerebrovascular accident"

PubMed

("Vagus Nerve Stimulation"[MeSH Terms] OR "Vagus Nerve Stimulation"[Title/Abstract] OR "Vagus Nerve"[Title/Abstract] OR "10th cranial nerve"[Title/Abstract] OR "cranial nerve X"[Title/Abstract] OR "non‐invasive vagus nerve"[Title/Abstract] OR "transcutaneous auricular vagus nerve"[Title/Abstract] OR "transcutaneous auricular nerve"[Title/Abstract] OR "VNS"[Title/Abstract] OR "ta‐VNS"[Title/Abstract] OR "pneumo‐gastric nerve"[Title/Abstract] OR "pneumo gastric nerve"[Title/Abstract] OR "pneumogastric nerve"[Title/Abstract] OR "stimulation of the vagus nerve"[Title/Abstract] OR "tenth cranial nerve"[Title/Abstract])

AND

("stroke"[MeSH Terms] OR "stroke, lacunar"[MeSH Terms] OR "embolic stroke"[MeSH Terms] OR "thrombotic stroke"[MeSH Terms] OR "stroke rehabilitation"[MeSH Terms] OR "stroke"[Title/Abstract] OR "cva"[Title/Abstract] OR "accident cerebrovascular"[Title/Abstract] OR "acute cerebrovascular lesion"[Title/Abstract] OR "acute stroke"[Title/Abstract] OR "brain vascular accident"[Title/Abstract] OR "cerebral stroke"[Title/Abstract] OR "cerebro vascular accident"[Title/Abstract] OR "cerebrovascular accident"[Title/Abstract]

Embase

('cerebrovascular accident'/exp OR 'cva':ti,ab,kw OR 'accident, cerebrovascular':ti,ab,kw OR 'acute cerebrovascular lesion':ti,ab,kw OR 'acute stroke':ti,ab,kw OR 'brain vascular accident':ti,ab,kw OR 'cerebral vascular accident':ti,ab,kw OR 'cerebral vascular insufficiency':ti,ab,kw OR 'cerebro vascular accident':ti,ab,kw OR 'cerebrovascular accident':ti,ab,kw OR 'cerebrum vascular accident':ti,ab,kw OR 'stroke':ti,ab,kw)

AND

('vagus nerve'/exp OR '10th cranial nerve' OR 'cranial nerve x' OR 'cranial nerve, tenth' OR 'nerve, tenth cranial' OR 'nerve, vagal' OR 'nervus vagus' OR 'pneumo‐gastric nerve' OR 'pneumogastric nerve' OR 'tenth cranial nerve' OR 'vagal nerve' OR 'vagus nerve' OR 'vagus nervus' OR 'vagus nerve stimulation'/exp OR 'vagal nerve stimulation':ti,ab,kw OR 'vagus nerve stimulation':ti,ab,kw OR 'implanted vagus nerve stimulator'/exp OR 't‐vns':ti,ab,kw OR 't‐vns stimulator cm02':ti,ab,kw OR 'vns therapy system':ti,ab,kw OR 'implantable vagus nerve stimulator':ti,ab,kw OR 'implanted vagus nerve stimulator':ti,ab,kw OR 'vagus nerve electrical stimulation system lead':ti,ab,kw OR 'vagus nerve electrical stimulation system programmer':ti,ab,kw OR 'vagus nerve electrical stimulators/stimulation systems':ti,ab,kw OR vns:ti,ab)

Scopus

(TITLE‐ABS‐KEY("Vagus Nerve Stimulation") OR TITLE‐ABS‐KEY("Vagus Nerve") OR TITLE‐ABS‐KEY("10th cranial nerve") OR TITLE‐ABS‐KEY("cranial nerve X") OR TITLE‐ABS‐KEY("non‐invasive vagus nerve") OR TITLE‐ABS‐KEY("transcutaneous auricular vagus nerve") OR TITLE‐ABS‐KEY("transcutaneous auricular nerve") OR TITLE‐ABS‐KEY("VNS") OR TITLE‐ABS‐KEY("ta‐VNS") OR TITLE‐ABS‐KEY("pneumo‐gastric nerve") OR TITLE‐ABS‐KEY("pneumo gastric nerve") OR TITLE‐ABS‐KEY("pneumogastric nerve") OR TITLE‐ABS‐KEY("stimulation of the vagus nerve") OR TITLE‐ABS‐KEY("tenth cranial nerve"))

AND

(TITLE‐ABS‐KEY("stroke") OR TITLE‐ABS‐KEY("cva") OR TITLE‐ABS‐KEY("accident cerebrovascular") OR TITLE‐ABS‐KEY("acute cerebrovascular lesion") OR TITLE‐ABS‐KEY("acute stroke") OR TITLE‐ABS‐KEY("brain vascular accident") OR TITLE‐ABS‐KEY("cerebral stroke") OR TITLE‐ABS‐KEY("cerebro vascular accident") OR TITLE‐ABS‐KEY("cerebrovascular accident"))

PsychINFO

Title/Abstract: "Vagus Nerve Stimulation" OR "Vagus Nerve" OR "10th cranial nerve" OR "cranial nerve X" OR "non‐invasive vagus nerve" OR "transcutaneous auricular vagus nerve" OR "transcutaneous auricular nerve" OR "VNS" OR "ta‐VNS" OR "pneumo‐gastric nerve" OR "pneumo gastric nerve" OR "pneumogastric nerve" OR "stimulation of the vagus nerve" OR "tenth cranial nerve"

AND

Title/Abstract: "stroke" OR "cva" OR "accident cerebrovascular" OR "acute cerebrovascular lesion" OR "acute stroke" OR "brain vascular accident" OR "cerebral stroke" OR "cerebro vascular accident" OR "cerebrovascular accident"

CINAHL

Abstract: "Vagus Nerve Stimulation" OR "Vagus Nerve" OR "10th cranial nerve" OR "cranial nerve X" OR "non‐invasive vagus nerve" OR "transcutaneous auricular vagus nerve" OR "transcutaneous auricular nerve" OR "VNS" OR "ta‐VNS" OR "pneumo‐gastric nerve" OR "pneumo gastric nerve" OR "pneumogastric nerve" OR "stimulation of the vagus nerve" OR "tenth cranial nerve"

AND

Abstract: "stroke" OR "cva" OR "accident cerebrovascular" OR "acute cerebrovascular lesion" OR "acute stroke" OR "brain vascular accident" OR "cerebral stroke" OR "cerebro vascular accident" OR "cerebrovascular accident"

PEDro

Abstract/Title: "Vagus Nerve Stimulation Stroke" (Matched any search term (OR))

OR Abstract/Title: "VNS Stroke" (Matched any search term (OR))

OR Abstract/Title: "Vagus Nerve Stimulation cva" (Matched any search term (OR))

OR Abstract/Title: "VNS cva" (Matched any search term (OR))

OR Abstract/Title: "Vagus Nerve Stimulation cerebrovascular accident" (Matched any search term (OR))

OR Abstract/Title: "VNS cerebrovascular accident" (Matched any search term (OR))

OR Abstract/Title: "Cranial Nerve Stroke" (Matched any search term (OR))

OR Abstract/Title: "Cranial Nerve cva" (Matched any search term (OR))

OR Abstract/Title: "Cranial Nerve cerebrovascular accident" (Matched any search term (OR))

Contributions of authors

GE created the search strategies under the supervision of GEB. All authors (GE, IS, JD, DET, SA, HMT, BOL and GEB) contributed to the conception and design, drafting, revision, and final approval of the version of the protocol manuscript to be published.

Sources of support

Internal sources

• School of Rehabilitation Sciences, Faculty of Medicine, Laval University, Canada

  Host institution (GE)

External sources

• Centre interdisciplinaire de recherche en réadaptation et intégration sociale, Canada

  Affiliated institution (GE)

• Neuro Musculo Skeletal Lab (NMSK), Institut de Recherche Expérimentale et Clinique, Secteur des Sciences de la Santé, UCLouvain, Belgium

  Affiliated Institution (GE)

• Wallonie‐Bruxelles International, Belgium

  Postdoctoral grant (GE)

• Réseau Provincial de Recherche en Adaptation et Réadaptation, Canada

  Postdoctoral grant (GE)

Declarations of interest

Gauthier Everard: none known

Ita Daryanti Saragih: works as a health professional in Kaohsiung Medical University (Taiwan)

Jesse Dawson: works as a health professional in NHS Greater Glasgow and Clyde. Received funding from MicroTransponder Inc since January 2012 for costs of enrolment for clinical trials and conference expenses. Investigator of several RCTs that are eligible for inclusion in this review (Dawson 2016; Dawson 2021; Kimberley 2018), and therefore, will not be involved in evaluating these studies. No other relevant conflict of interests.

Dame Elysabeth Tarihoran: none known

Shailesh Advani: none known

Huey‐Ming Tzeng: registered nurse at the University of Texas Medical Branch (Galveston, TX, USA)

Bih‐O Lee: Professor and Dean of College of Nursing of Kaohsiung Medical University (Taiwan)

Geertruida Bekkering: none known

New