Neurostimulation in People with Oropharyngeal Dysphagia: A Systematic Review and Meta-Analysis of Randomised Controlled Trials—Part II: Brain Neurostimulation

Objective. To assess the effects of brain neurostimulation (i.e., repetitive transcranial magnetic stimulation [rTMS] and transcranial direct current stimulation [tDCS]) in people with oropharyngeal dysphagia (OD). Methods. Systematic literature searches were conducted in four electronic databases (CINAHL, Embase, PsycINFO, and PubMed) to retrieve randomised controlled trials (RCTs) only. Using the Revised Cochrane risk-of-bias tool for randomised trials (RoB 2), the methodological quality of included studies was evaluated, after which meta-analysis was conducted using a random-effects model. Results. In total, 24 studies reporting on brain neurostimulation were included: 11 studies on rTMS, 9 studies on tDCS, and 4 studies on combined neurostimulation interventions. Overall, within-group meta-analysis and between-group analysis for rTMS identified significant large and small effects in favour of stimulation, respectively. For tDCS, overall within-group analysis and between-group analysis identified significant large and moderate effects in favour of stimulation, respectively. Conclusion. Both rTMS and tDCS show promising effects in people with oropharyngeal dysphagia. However, comparisons between studies were challenging due to high heterogeneity in stimulation protocols and experimental parameters, potential moderators, and inconsistent methodological reporting. Generalisations of meta-analyses need to be interpreted with care. Future research should include large RCTs using standard protocols and reporting guidelines as achieved by international consensus.


Introduction
Oropharyngeal dysphagia (OD) or swallowing problems is highly prevalent among stroke patients, people with progressive neurological diseases, patients with head and neck cancer, and in frail older persons [1,2]. Prevalence estimates of OD may vary depending on underlying medical diagnoses, but have been reported as high as 80% in stroke and Parkinson's disease [3], and 70% in oncological populations [4]. OD is associated with dehydration, malnutrition, aspiration pneumonia, and increased mortality [5][6][7], but also leads to decreased health-related quality of life [8].
Treatment and management of OD may vary widely. However, apart from traditional compensatory and rehabilitative strategies including diet modifications, postural adjustments, oromotor training and swallow manoeuvres [9], recent studies report on the possible beneficial effects of non-invasive brain stimulation. Brain neurostimulation aims to modulate cortical excitability and include techniques such as repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS). rTMS uses electromagnetic induction resulting in depolarisation of postsynaptic connections, whereas tDCS uses direct electrical current shifting the polarity of nerve cells [10]. Neurostimulation protocols may vary greatly per study, including different neurostimulation sites, frequencies, stimulation duration and number of different outcome measures are used to objectify treatment effects, and individual responses to stimulation are highly variable [10][11][12].
Aspiring to improved treatment efficacy in OD management, non-invasive brain stimulation has achieved growing interest over the past decade. Several reviews have been published on rTMS and tDCS [10,[12][13][14][15][16][17][18], each publication having different inclusion and exclusion criteria and methodology. All previous reviews targeted brain neurostimulation interventions in post-stroke populations except for one review that included patients with acquired brain injury [16]; to date, all reviews on brain stimulation set criteria based on medical diagnoses. Moreover, not all reviews performed meta-analysis [14] and as several neurostimulation trials have only been published recently, earlier reviews will have identified fewer studies. This is the second paper (Part II) of two companion papers on treatment effects of neurostimulation in people with OD. The first systematic review (Part I) reported on the effects of pharyngeal electrical stimulation (PES) and neuromuscular electrical stimulation (NMES).
The aim of this systematic review (Part II) is to determine the effects of brain neurostimulation (i.e., rTMS and tDCS) in people with OD without excluding populations based on medical diagnoses. Only randomised controlled trials (RCTs) will be included being the highest level of evidence. Meta-analyses will be conducted to summarise results and report on possible moderators of treatment effects.

Inclusion and Exclusion Criteria
To be eligible for inclusion in this systematic review, studies had to meet the following criteria: (1) participants had a diagnosis of oropharyngeal dysphagia; (2) the study included non-invasive neurostimulation interventions aimed at reducing swallowing or feeding problems; (3) the study included a control group or comparison intervention group; (4) participants were randomly assigned to one of the study arms or groups; and (5) the study was published in English language.
Interventions such as non-electrical peripheral stimulation (e.g., air-puff or gustatory stimulation), pharmacological interventions and acupuncture, were considered out of scope of this review, thus were excluded. Invasive techniques and/or those that did not specifically target OD (e.g., deep-brain stimulation studies after neurosurgical implementation of a neurostimulator) were also excluded. Conference abstracts, doctoral theses, editorials, and reviews were excluded.

Systematic Review
Methodological Quality and Risk of Bias. The Revised Cochrane risk-of-bias tool for randomised trials (RoB 2) [21] was used to assess the methodological quality of the included studies. The RoB 2 tool identifies domains to consider when assessing where bias may have been introduced into a randomised trial: (1) bias arising from the randomisation process; (2) bias due to deviations from intended interventions; (3) bias due to missing outcome data; (4) bias in measurement of the outcome; and (5) bias in selection of the reported result. For each domain, a series of signalling questions are answered to give a judgement (i.e., "low risk of bias", "some concerns", or "high risk of bias"), which can then be assessed in aggregate to determine a study's overall risk of bias [21].
Data Collection Process. Data were extracted from the included studies using a data extraction form created for this purpose. This form allowed for extraction of data under several categories, relevant to meta-analyses, including participant diagnosis, inclusion and exclusion criteria, sample size, age, gender, intervention goal, intervention agent/delivery/dosage, outcome measures, and treatment outcomes.
Data, Items and Synthesis of Results. Titles and abstracts of included studies were reviewed for eligibility by two independent reviewers. Next, the same two reviewers assessed the selected original articles at a full-text level to determine their eligibility. To ensure rating accuracy, a random selection of one hundred records were scored and discussed over two consecutive group sessions prior to rating the remaining records. Any disagreement Utilising Comprehensive Data Analysis software, publication bias was evaluated as per the Begg and Muzumdar's rank correlation test and the Fail-safe N test. Begg and Muzumdar's rank correlation test provides information on the rank correlations between standardised effect size and the ranks of their variances [25]. In addition to a tau value, a two-tailed p value is also generated. Where the analysis results in a value of zero, it can be concluded that there is unlikely to be an association between the effect size and ranks of variance. Conversely, the closer to one the tau or p values, the more likely there is to be an association between the effect size and ranks of variance. Therefore, high standard error would be connected to higher effect sizes if publication bias was the result of asymmetry. If larger effects are represented by low values, tau would be over zero; conversely tau would be negative if larger effects are represented by high values.
The Fail-safe N test is a calculation of the quantity of studies with zero effect size that could be incorporated into the meta-analysis prior to the result losing statistical significance, that is, the quantity of excluded studies that would result in the effect being nullified [26]. Results should be treated with care where the fail-safe N is relatively small, however, when it is large, conclusions can be confidently drawn that the treatment effect, while potentially raised by the removal of some studies, is not nil.

Study Selection
A total of 8059 studies were retrieved through the subject heading and free text searches (CINAHL: n = 239, Embase: n = 4550, PsycINFO: n = 231, and PubMed: n = 3039). Following removal of duplicates at a title and abstract level (n = 1113), a total of 6946 records remained. A total of 261 original articles were assessed at a full-text level, with articles grouped according to type of intervention. At this stage, no studies were excluded based on type of intervention (e.g., behavioural intervention, neurostimulation). Of these, 58 articles on neurostimulation were identified that satisfied the inclusion criteria. Four additional studies were found through reference checking of the included articles. This process resulted in a final number of 24 included studies. Figure 1 presents the flow diagram of the overall reviewing process according to PRISMA. Utilising Comprehensive Data Analysis software, publication bias was evaluated as per the Begg and Muzumdar's rank correlation test and the Fail-safe N test. Begg and Muzumdar's rank correlation test provides information on the rank correlations between standardised effect size and the ranks of their variances [25]. In addition to a tau value, a two-tailed p value is also generated. Where the analysis results in a value of zero, it can be concluded that there is unlikely to be an association between the effect size and ranks of variance. Conversely, the closer to one the tau or p values, the more likely there is to be an association between the effect size and ranks of variance. Therefore, high standard error would be connected to higher effect sizes if publication bias was the result of asymmetry. If larger effects are represented by low values, tau would be over zero; conversely tau would be negative if larger effects are represented by high values.
The Fail-safe N test is a calculation of the quantity of studies with zero effect size that could be incorporated into the meta-analysis prior to the result losing statistical significance, that is, the quantity of excluded studies that would result in the effect being nullified [26]. Results should be treated with care where the fail-safe N is relatively small, however, when it is large, conclusions can be confidently drawn that the treatment effect, while potentially raised by the removal of some studies, is not nil.

Study Selection
A total of 8059 studies were retrieved through the subject heading and free text searches (CINAHL: n = 239, Embase: n = 4550, PsycINFO: n = 231, and PubMed: n = 3039). Following removal of duplicates at a title and abstract level (n = 1113), a total of 6946 records remained. A total of 261 original articles were assessed at a full-text level, with articles grouped according to type of intervention. At this stage, no studies were excluded based on type of intervention (e.g., behavioural intervention, neurostimulation). Of these, 58 articles on neurostimulation were identified that satisfied the inclusion criteria. Four additional studies were found through reference checking of the included articles. This process resulted in a final number of 24 included studies. Figure 1 presents the flow diagram of the overall reviewing process according to PRISMA.    Tables 2 and 3 report detailed descriptions of all included studies. Table 2 includes data on study characteristics including methodological study quality, inclusion and exclusion criteria, and details on participant groups. Information is provided for all study groups (control and intervention groups), medical diagnosis, sample size, age and gender. Table 3 reports on intervention characteristics, including goals, intervention components, outcome measures, intervention outcomes, as well as main conclusions.

Description of Studies
Brain stimulation Interventions (Table 2). Across the 24 included studies, eleven studies reported on rTMS and nine studies reported on tDCS. Four studies used another type of neurostimulation (i.e., NMES) in addition to rTMS, either within the same group or over different treatment groups.
Outcome Measures (Table 3). Outcomes measures varied greatly across all studies included in the review, covering several domains within the area of OD. The Penetration Aspiration Score (PAS) was the most reported outcome measure (8 studies), followed by the Dysphagia Outcome and Severity Scale (DOSS; 7 studies), Functional Oral Intake Scale (FOIS; 3 studies) and Degree of Dysphagia (DD; 3 studies).
rTMS Intervention (n = 11: Tables 2 and 3). All but one of the rTMs studies [33] compared rTMS stimulation with sham rTMS. One single study compared rTMS with rTMS combined with DT, and DT only [33]. Three more studies included three arms; two studies compared rTMS using different frequencies versus sham rTMS [32,34], and one study compared bilateral and unilateral rTMS versus sham rTMS [35].
Combined Neurostimulation Interventions (n = 4: see Tables 2 and 3). Three studies in the combined intervention group compared three different treatments. Of these, one compared rTMS, PES and paired associative stimulation (PAS) [42], a second compared DT, rTMS combined with DT, and NMES combined with DT [43], and a third compared rTMS, PES and capsaicin stimulation [44]. A fourth study combined NMES stimulation with sham rTMS or rTMS stimulating different hemispheres (ipsilesional, contralesional or bilateral) [45].   Figure-of-eight coil placed over oesophageal cortical area of both hemispheres, judged to be about 3 cm anterior and 6 cm lateral to the vertex (neurophysiology explorations not performed on participants due to severity of vertigo and dysphagia).
Sham group: • Similar parameters producing the same noise, but with coil rotated away from scalp • Similar parameters producing the same noise, but with the coil rotated away from scalp • Similar parameters to high frequency stimulation producing the same noise, but with the coil rotated away from scalp             Treatment group 2: Ipsilateral rTMS + NMES 10-Hz real rTMS was delivered to the hot spot for the mylohyoid muscle at the ipsilesional hemisphere followed by 1-Hz sham rTMS over the corresponding position of the contralesional hemisphere. Treatment group 3: Contralateral rTMS + NMES 10-Hz sham rTMS was delivered to the hot spot for the mylohyoid muscle at the ipsilesional hemisphere followed by 1-Hz real rTMS over the corresponding position of the contralesional hemisphere. Treatment group 4: Bilateral rTMS + NMES 10-Hz real rTMS was delivered to the hot spot for the mylohyoid muscle at the ipsilesional hemisphere followed by 1-Hz real rTMS over the corresponding position of the contralesional hemisphere.
a Where information was available on how stimulation site was located and mapped, and whether stimulation was applied ipsilateral or contralateral to the lesion site, it was included.  To investigate the short-(2-months) and long-term ( To investigate the effects of high-frequency versus low-frequency rTMS on poststroke dysphagia during early rehabilitation Primary outcome: SSA [53]. Secondary outcomes: WST [54], DD [55], NIHSS score [56], BI [57], mRS, measures of mylohyoid MEPs evoked from both hemispheres before and after treatment. Assessed: before treatment, after 5th rTMS session, and at 1-, 2-, and 3-months post-treatment.

•
FDS and PAS improved significantly in the low intensity group compared to other groups • Significant improvement in ASHA-NOMS Swallow Scale in the sham and low intensity groups   [35] to investigate the effects of high-frequency rTMS at the bilateral motor cortices over the cortical representation of the mylohyoid muscles in the patients with post-stroke dysphagia.
Primary outcomes: Immediately post-treatment and 3 weeks post-treatment: using CDS [69], DOSS [58], PAS [66], and VDS [68].   To identify whether applying low-frequency rTMS can enhance the effect of conventional swallowing treatment and quality of life of chronic (2-6 months) stroke patients suffering from dysphagia Primary outcome: PAS [66], pre-post treatment, 1 and 3 months post-treatment. Secondary outcomes: VFSS parameters (including oral parameters, tongue retraction, hyolaryngeal elevation, delayed swallow reflex, residue, nutritional status, SWAL-QOL).   To evaluate the efficacy of a pathophysiologically reasonable tDCS protocol to improve stroke-related OD, via a randomized controlled trial (RCT) in a sufficiently large patient sample with objective clinical outcome measures alongside functional neuroimaging   [40] To investigate the effects of tDCS combined with conventional swallowing training on the swallowing function in brainstem stroke patients with cricopharyngeal muscle dysfunction.
Primary outcomes: Statistical difference between the groups at endpoint not reported. • Secondary outcomes showed no significant differences between the groups.

Combined Neurostimulation Interventions-n = 4
Cabib et al. (2020) [44] To investigate the effect of rTMS of the primary sensory cortex (A), oral capsaicin (B) and intra-pharyngeal electrical stimulation (IPES; C) on post-stroke dysphagia Primary outcomes: Effect size pre-post treatment for neurophysiological variables (pharyngeal and thenar RMT and MEP). Secondary outcomes: Effects on the biomechanics of swallow (PAS [66], impaired efficiency + more) VFSS before and after treatment   [43] To investigate the effect of low-frequency rTMS and NMES on post-stroke dysphagia.  To determine whether rTMS NMES effectively ameliorates dysphagia and how rTMS protocols (bilateral vs. unilateral) combined with NMES can be optimized.
Compared with group 2 or 3 in the affected hemisphere, group 4 displayed a significantly greater percentage change (p.0.017 and p.0.024, respectively). All groups displayed significant improvements in SSA and DD scores after treatment and at 1-month follow-up. The percentage change in cortical excitability increased over time in either the affected or unaffected hemisphere in treatment groups 1, 2 and 4 (p < 0.05). In Group 3, the percentage change in cortical excitability in the unaffected hemisphere significantly decreased after the stimulation course (p < 0.05). Change in SSA and DD scores in group 4 was markedly higher than that in the other three groups at the end of stimulation (p.0.02, p.0.03, and p.0.005) and still higher than that in group 1 at the 1-month follow-up (p.0.01).

Risk of Bias Assessment and Methodological Quality
The Begg and Mazumdar rank correlation procedure produced a tau of −0.036 (twotailed p = 0.902) and 0.178 (two-tailed p = 0.536) for rTMS and tDCS, respectively. The rTMS meta-analysis incorporates data from 8 studies, which yield a z-value of 2.348 (two-tailed p-value = 0.019). The fail-safe N is 4. This means that 4 'null' studies need to be located and included for the combined two-tailed p-value to exceed 0.050. That means there would be need to be 0.5 missing studies for every observed study for the effect to be nullified. The tDCS meta-analysis incorporates data from 8 studies yielding a z-value of 4.857 (two-tailed p-value < 0.001). The fail-safe N is 42 indicating 42 'null' studies need to be located and included for the combined two-tailed p-value to exceed 0.050; there would be need to be 5.3 missing studies for every observed study for the effect to be nullified. Both of these procedures (i.e., Begg and Mazumdar rank correlation and fail-safe N test) indicate the absence of publication bias. Figures 2 and 3 present, respectively, the risk of bias summary per domain for all included studies combined and for individual studies, assessed using the Revised Cochrane Collaboration tool for assessing risk of bias (RoB 2) [21]. The majority of studies had low risk of bias with very few exceptions.

Risk of Bias Assessment and Methodological Quality
The Begg and Mazumdar rank correlation procedure produced a tau of −0.036 (twotailed p = 0.902) and 0.178 (two-tailed p = 0.536) for rTMS and tDCS, respectively. The rTMS meta-analysis incorporates data from 8 studies, which yield a z-value of 2.348 (two-tailed p-value = 0.019). The fail-safe N is 4. This means that 4 'null' studies need to be located and included for the combined two-tailed p-value to exceed 0.050. That means there would be need to be 0.5 missing studies for every observed study for the effect to be nullified. The tDCS meta-analysis incorporates data from 8 studies yielding a z-value of 4.857 (two-tailed p-value < 0.001). The fail-safe N is 42 indicating 42 'null' studies need to be located and included for the combined two-tailed p-value to exceed 0.050; there would be need to be 5.3 missing studies for every observed study for the effect to be nullified. Both of these procedures (i.e., Begg and Mazumdar rank correlation and fail-safe N test) indicate the absence of publication bias. Figures 2 and 3 present, respectively, the risk of bias summary per domain for all included studies combined and for individual studies, assessed using the Revised Cochrane Collaboration tool for assessing risk of bias (RoB 2) [21]. The majority of studies had low risk of bias with very few exceptions.

rTMS Meta-Analysis
Eight studies using rTMS [32,33,35,[42][43][44]50,51] were included in the meta-analysis. Of these, three studies provided data for two different interventions groups [32,35,36]. Six studies were excluded as OD was not confirmed by instrumental assessment and one study was excluded as rTMS was combined with NMES.
Overall within-group analysis. Pre-post intervention effect sizes ranged from 0.085 to 2.068 ( Figure 4) with seven studies showing large effect sizes (Hedges' g > 0.8). Pre-post interventions produced a significant, large effect size (Hedges' g = 1.038).  [21]. Note. If one or more yellow circles (domains) have been identified for a particular study, the Overall score (last column) shows an exclamation mark, indicating that the study shows some concerns (yellow circle with exclamation mark).
Overall within-group analysis. Pre-post intervention effect sizes ranged from 0.085 to 2.068 ( Figure 4) with seven studies showing large effect sizes (Hedges' g > 0.8). Pre-post interventions produced a significant, large effect size (Hedges' g = 1.038). Overall between-group analysis. A significant, small post-intervention betweengroup total effect size was calculated in favour of rTMS (random-effects model: z(7) = 2.338, p = 0.019, Hedges' g = 0.355, and 95% CI = 0.057-0.652; Figure 5). Between-study heterogeneity was non-significant (Q(7) = 6.763, p = 0.454). Between-subgroup analyses. Subgroup analyses were conducted to compare time between pre-and post-intervention measurement, stimulation sites (bilateral, contra-lesional and ipsi-lesional sites), pulse ranges (low: ≤600; medium; >600 and <10,000; high: ≥10,000 pulses), stimulation frequencies (1, 5 and 10 Hz), and optional behavioural training (rTMS versus rTMS + DT;  Between-subgroup analyses. Subgroup analyses were conducted to compare time between pre-and post-intervention measurement, stimulation sites (bilateral, contralesional and ipsi-lesional sites), pulse ranges (low: ≤600; medium; >600 and <10,000; high: ≥10,000 pulses), stimulation frequencies (1, 5 and 10 Hz), and optional behavioural training (rTMS versus rTMS + DT; Table 4). No subgroup comparisons for outcome measures were conducted as all but one study used PAS. Studies including a longer time span between pre-and post-interventions (indicating longer stimulation times) showed increased positive effect sizes compared to one-day interventions, which showed negligible effect sizes. When comparing stimulation sites, non-significant, positive effect sizes were obtained for all three stimulation groups with large ranges in effect sizes within groups.
Pulse range comparisons indicated an increased significant, positive effect for higher pulse ranges. Effect sizes were only significant for large numbers of pulses delivered. Subanalyses comparing stimulation frequencies did not indicate obvious tendencies between groups. rTMS in combination with DT showed non-significant, small positive effect sizes in one study, whereas DT alone showed similar significant, small effects sizes.
Overall within-group analysis. The overall pre-post intervention effect size was 1.385, with effect sizes ranging from 0.432 (small effect) to 3.365 (high effect; Figure 6). Studies showed small (n = 2), moderate (n = 1), and high effect sizes (n = 5).
Between subgroup analyses. Subgroup analyses were conducted comparing time between pre-and post-intervention measurements, outcome measures, total stimulation times and stimulation current (Table 4). Increasing the number of days between pre-and post-intervention showed a strong tendency towards increased positive effect sizes, with significant effect sizes for two and four-week periods. Comparisons between measures resulted in significant, large positive effect sizes for visuoperceptual evaluation of  Between subgroup analyses. Subgroup analyses were conducted comparing time between pre-and post-intervention measurements, outcome measures, total stimulation times and stimulation current (Table 4). Increasing the number of days between pre-and post-intervention showed a strong tendency towards increased positive effect sizes, with significant effect sizes for two and four-week periods. Comparisons between measures resulted in significant, large positive effect sizes for visuoperceptual evaluation of instrumental assessment, but negligible effects when using an oral intake measure. Effect sizes for comparisons between total stimulation times indicated increased effects when using longer stimulation times. Significant, large effects were demonstrated for stimulation times of 300 min and longer. Additionally, higher stimulation currents resulted in increased significant, large positive effect sizes.

Discussion
This systematic review (Part II) aimed to determine the effects of rTMS and tDCS in people with OD. This systematic review and meta-analysis of RCT studies were completed in accordance with PRISMA procedures [19,20]. No populations were excluded based on medical diagnoses.

Systematic Review Findings
Like the systematic review on effects of NMES and PES in people with OD (Part I) [75], methodological problems were identified relating to unclear definitions of OD and differences in methods of confirming the presence of OD (i.e., using instrumental assessment, patient self-report or clinical assessment). Consequently, to reduce heterogeneity in participant characteristics between RCTs, only studies using instrumental assessment to confirm diagnosis of OD were included in meta-analyses. As most studies included stroke patients only, no meta-analysis could be performed to determine effects per medical diagnosis.
With the exception of one study [33], all rTMS studies included in the meta-analysis used the PAS to evaluate intervention effects. For the tDCS studies, as heterogeneity in outcome measures was larger, data on three different clinical outcome measures were used when conducting the meta-analysis. All rTMS studies used sham stimulation as a comparison group with the exception of one study which included a rTMS plus DT group [33]. For the tDCS studies, all but one study [31] combined neurostimulation with simultaneous DT. When comparing the degree of heterogeneity in study designs between brain neurostimulation (i.e., rTMS and tDCS) and peripheral neurostimulation (i.e., NMES and PES), those in the peripheral neurostimulation group were more diverse, creating greater challenges for conducting meta-analyses. Non-invasive brain stimulation studies tended to recruit smaller sample populations compared to peripheral studies [75].

rTMS
This review prioritised reducing heterogeneity for purposes of meta-analysis. In contrast to previously published reviews that did not confirm OD by instrumental assessment, those studies were excluded from this meta-analysis. With the exception of Bath, Lee [13], earlier reviews identified significant beneficial effects of rTMS. Therefore, even though comparing the current meta-analysis with analyses from previous reviews may be challenging due to the inclusion of different outcome data, the findings from these studies seem in line with each other and this review.

tDCS
Fewer RCTs were identified for tDCS compared with rTMS. Eight out of nine studies were eligible for meta-analysis, with one study excluded due to insufficient data; this was the only study to include non-stroke patients (presbydysphagia) [31]. Again, as previous reviews on tDCS [10,12,13,[16][17][18] applied different criteria for inclusion and study methodology (e.g., differences in selection of electronic databases and publication years), final numbers of studies used for these meta-analyses ranged between two and seven publications, with reviews published before 2020 including four or fewer studies. When comparing the present results with the two most recent reviews [10,18] (both including seven studies), the beneficial effects of tDCS identified by this review were confirmed by significant, small-to-moderate effects in favour of tDCS.

Moderators
Several factors may have had an impact on conducting meta-analyses and results. Comparing previous reviews, different decisions were made concerning criteria for metaanalyses. For example, Bath, Lee [13] excluded comparison groups with active treatment components and Chiang, Lin [12] excluded chronic stroke patients. Chronicity of stroke has shown to influence effect sizes [10,18], but selecting different primary outcomes may also result in deviating findings. For instance, Bath, Lee [13] did not find any positive effects for either rTMS or tDCS on primary outcome measures defined as death or dependency at the end of trials. Additionally, underlying medical diagnoses of OD are expected to affect meta-analyses. However, no conclusions could be drawn as very few studies of non-stroke patients were included in this review, thus no meta-analysis differentiating between diagnoses was conducted.
Similar reasons for hindering comparisons between RCTs are present in the current review, for example, spontaneous recovery and stroke severity, as were identified in the systematic review on effects of NMES and PES in people with OD (Part I) [75]. To account for the possibility of spontaneous recovery in participants, only between-subgroup metaanalyses were conducted using post-intervention data. However, the effects of stroke severity linked to OD severity remains unclear as RCTs usually did not report on the severity of stroke in sufficient detail.
Lastly, brain neurostimulation between RCTs may differ with respect to stimulation protocols (e.g., stimulation site, number and duration of treatment sessions and period) and technical parameters (e.g., frequency or number of pulses). The relatively low numbers of RCTs included in this review meant that meta-analysis could not incorporate all potential moderators. However, many of the included studies lacked sufficient details on technical parameters to allow further comparisons.

Limitations
Although this review followed PRISMA guidelines and aimed at reducing bias, some limitations may have had an impact on the results as presented. Only RCTs published in English were eligible in this review. Thus, some RCTs may have been excluded based on language criteria when their findings could have contributed to the current meta-analysis. Moreover, the high degree of heterogeneity between included studies hampered meta-analyses. Therefore, the results of meta-analyses and generalisations made should be interpreted with care.

Conclusions
The results of this systematic review suggest that both rTMS and tDCS show promising effects in people with OD. Meta-analysis for RCTs identified large pre-post intervention effect sizes for both types of brain neurostimulation. In addition, this analysis found significant, small and moderate post-intervention between-group effects in favour of rTMS and tCDS, respectively. However, comparisons between studies remain uncertain and challenging due to high heterogeneity in stimulation protocols and experimental parameters, potential moderators of stimulation effects, small samples sizes, and inconsistent methodological reporting.
These findings suggest that there is a need for RCTs including larger sample sizes to support future meta-analyses that will be able to adequately account for the presence of moderators. In addition, international consensus on standardised study protocols and reporting guidelines is required to support comparisons between studies.