Using non-invasive transcranial stimulation to improve motor and cognitive function in Parkinson’s disease: a systematic review and meta-analysis

Parkinson's disease (PD) is a neurodegenerative disorder affecting motor and cognitive abilities. There is no cure for PD, therefore identifying safe therapies to alleviate symptoms remains a priority. This meta-analysis quantified the effectiveness of repetitive transcranial magnetic stimulation (rTMS) and transcranial electrical stimulation (TES) to improve motor and cognitive dysfunction in PD. PubMed, EMBASE, Web of Science, Google Scholar, Scopus, Library of Congress and Cochrane library were searched. 24 rTMS and 9 TES studies (n = 33) with a sham control group were included for analyses. The Physiotherapy Evidence Database and Cochrane Risk of Bias showed high quality (7.5/10) and low bias with included studies respectively. Our results showed an overall positive effect in favour of rTMS (SMD = 0.394, CI [0.106–0.683], p = 0.007) and TES (SMD = 0.611, CI [0.188–1.035], p = 0.005) compared with sham stimulation on motor function, with no significant differences detected between rTMS and TES (Q [1] = 0.69, p = 0.406). Neither rTMS nor TES improved cognition. No effects for stimulation parameters on motor or cognitive function were observed. To enhance the clinical utility of non-invasive brain stimulation (NBS), individual prescription of stimulation parameters based upon symptomology and resting excitability state should be a priority of future research.

may diminish over time 9 and lead to compulsive behaviours 10 . Furthermore, long-term use of anticholinergic medications can increase the risk of dementia 11 , and in addition to surgical risks, DBS can induce mood, memory and personality disturbances 12 . On this basis, non-invasive brain stimulation (NBS) has been identified as an alternative therapy, with the potential to alleviate both the motor and cognitive symptoms and minimise risks associated with traditional treatment options.
The use of NBS, which includes repetitive transcranial magnetic stimulation (rTMS), and transcranial electric stimulation (TES) paradigms such as transcranial direct current stimulation (tDCS) and transcranial alternating current stimulation (tACS) 13 , have gained significant interest as an adjunct treatment for PD. While the mechanisms of how rTMS and TES modulate cortical excitability differ (rTMS induces direct and trans-synaptic neuronal activation 14 , while TES induces subthreshold neuronal membrane polarization 15,16 ), both methods have been shown to induce long-term potentiation or depression (LTP/LTD)-like synaptic plasticity [17][18][19][20] . Given the alterations in cortico-striatal excitatory and inhibitory transmission observed in PD, these NBS techniques may serve to normalise the aberrant neurophysiology associated with PD by modulating the excitability of underlying cortical tissue.
The physiological response from NBS can be manipulated through different stimulation parameters. High-frequency rTMS (greater than 5 Hz) and anodal-tDCS typically increase cortical excitability whilst low-frequency rTMS (less than 1 Hz) and cathodal-tDCS result in the opposite effect 21,22 . Differential outcomes from NBS have also been attributed to the timing, duration, electrode placement, coil orientation, and the physiological state of the participant 23 . While manipulation of stimulation parameters offers the potential for individualising treatment, it also highlights the importance of understanding the magnitude of effect from different NBS protocols.
Previous reviews in PD have supported overall improvements in motor function [24][25][26][27] , but not other subsections of the Unified Parkinson's disease rating scale (UPDRS) 27 . Support for differential effects of TMS parameters (frequency and no. of pulses) have been highlighted in rTMS reviews 24,28 , however the optimal tDCS parameters have not been comprehensively examined. Moreover, no review has compared the efficacy of the different stimulation modes (TES and rTMS) on both motor and cognitive outcomes. This is important as TES is simple to apply, more portable and relatively inexpensive compared with rTMS, which may serve as a more clinically-viable tool for routine use in PD therapy. This is the first review to compare the efficacy of two commonly applied NBS techniques on both motor performance (as measured by the unified PD rating scale [UPDRS III], gait and hand movement), and cognitive performance (executive functions, psychomotor speed, visuospatial abilities and memory) in people with PD. We also examine whether different stimulation parameters modulate these outcomes, and whether the optimal stimulation parameters are different for motor and cognitive symptoms.
Five TES studies used 2 mA 55,57,59-61 , one used 1 mA 58 , while the other three studies used both 1 and 2 mA intensities on separate sessions 53,54,56 . Only two different stimulation sites were used amongst TES studies, which were the M1 53,58,59,61 , PFC 55,57 or both 54,56,60 . To achieve sham stimulation, TES devices had an in-built sham function providing a short ramp-up and down effect, allowing for pseudo-stimulation. Table 3) and Cochrane Risk of Bias tool (Supplementary Table 4) showed high internal validity (7.5/10) and a low risk of bias. The main sources of bias reported from both the PEDro and Cochrane risk tools were lack of concealed allocation, unclear randomisation/sequence generation procedures and lack of assessor blinding. While the lack of concealed allocation and assessor blinding may have potentially biased the findings of the studies, our analyses of publication bias suggest otherwise. However, it should be noted that due to the lack of rTMS and TES studies on cognitive function ( Fig. 5B and D), caution is needed when interpreting the publication bias results.   Number of sessions to predict effect size. Meta-regression analysis showed that number of session was not a strong predictor of motor function outcomes following rTMS (R 2 = 0.042, p = 0.298). Meta-regression of rTMS and TES on cognitive function and TES on motor function were excluded due to the low number of studies (≤5 studies) available to power the analysis.

Discussion
This systematic review and meta-analysis included 33 studies examining the effects of rTMS and TES on motor and cognitive symptoms in individuals with PD. Building on evidence from previous meta-analyses, our study (i) provided a direct comparison between the effect sizes of rTMS and TES studies; (ii) compared the effects of rTMS and TES on both motor and cognitive function; and (iii) included a larger pool of studies than previous reviews. The results from our meta-analyses support the use of rTMS and TES to improve motor deficits in PD. Although mixed-effect analyses showed no differences in motor function between TES and rTMS, when examining specific motor domains, we showed a more widespread effect for rTMS across both UPDRS-III scores and gait parameters. However our understanding as to the benefit of NBS to improve cognitive function is limited by the smaller (5 rTMS and 3 TES) number of studies. The subgroup and meta-regression showed no differential effects of rTMS and TES moderator variables on motor or cognitive function.
Our findings suggest that rTMS of the M1 and PFC elicited the strongest effects on motor function. This agreed with a previous meta-analysis by Chou, et al. 24 , which showed that high-frequency (≥5 Hz) rTMS over the M1 and low-frequency (≤1 Hz) over the prefrontal areas, led to greater improvements in motor function as measured by the UPDRS III. The rationale for using high-frequency stimulation over the motor areas (i.e. M1 and SMA) appears logical, given that PD is associated with hypo-activity of these areas during self-initiated movements 62 . However, our results suggest no difference between high-and low-frequency rTMS overall, which may be in part due to large variability in results from existing studies. For example, several studies that applied 5 Hz rTMS over the SMA demonstrated significant improvements in bradykinesia 35 and handwriting performance 49 , however Boylan, et al. 51 showed that high frequency (10 Hz at 110%MT) rTMS of the SMA worsened performance in both fine and complex movements. It has been suggested that higher frequencies applied over the SMA may be counterproductive due to homeostatic plasticity 63 caused by hyperactivity of the SMA at rest 40,64 . Of interest, our results further showed that rTMS over the PFC may improve motor function as well. While the number of studies investigating the role of the PFC in motor function in PD is limited, it is possible that stimulating the PFC increases endogenous dopamine production from the basal ganglia via corticostriatal pathways 65,66 , which may serve to alleviate motor deficits. Based on the current evidence, the effects of rTMS seem to be largely site-specific (i.e. M1, SMA or PFC) and dependent on the outcome measures to be improved. Due to the inherently large variability in rTMS responses in PD 67 , the clinical utility of rTMS might benefit from individually-tailored parameters, to more effectively manage cognitive or motor impairments in PD.
Our results from TES studies indicate a significant positive effect for measures of gait performance but not for measures of hand movement (albeit approaching significance, P = 0.065) and UPDRS III. Furthermore, our comparison between the overall pooled effects of rTMS and TES showed no significant difference in motor improvements. While the specific mechanisms by which TES improves motor function in people with PD are unclear, a proposed mechanism is that cortical stimulation may have far-reaching effects on subcortical brain structures such as the basal ganglia, via indirect cortico-subcortical projections 68 . Indeed a recent study in humans showed that the effects of TES could extend to subcortical structures 69 and potentially normalise dysfunctional inhibitory circuits within the basal-thalamic-cortico pathways 18,20 . Based on our findings, the fact that the TES pooled effect in motor improvements was similar to rTMS may therefore have greater clinical relevance due to its relatively inexpensive and portable nature compared with rTMS systems 70 . In this context, TES can also be applied concurrently or as a priming technique with physical therapy 71,72 . It is plausible that this application may reinforce LTP-like processes 73,74 , promoting greater retention of gains from physical therapy 75,76 . Although beyond the scope of this review, comparisons of TES applied in isolation or with physical therapy on longer-term management of motor symptoms should be a focus of future research.
In line with a previous systematic review by Broeder,et al. 26 , our overall pooled results suggest that TES had a greater effect on motor function and to a lesser extent on cognitive function in people with PD. This is likely to be due to the inherent nature of most neuropsychological tests used (i.e., TMT and reaction time tests) in our studies, which consists of motor and cognitive components that may be affected differently by TES. In this sense, it is difficult to discern if the improvements in cognitive test scores were indeed due to better cognitive functioning following TES, or whether TES had improved motor abilities that would result in better overall test scores. While the findings from individual studies within our search seem to indicate stronger effects on working memory using anodal tDCS at a higher stimulation intensity (2 mA) 54,56 , and over the left DLPFC, more sensitive cognitive tests are required to discriminate between the effects of TES on cognitive and motor abilities. Furthermore, cognitive processing is likely to be underpinned by several brain regions and neural networks, which makes it difficult to identify specific regions to stimulate. This is is unlike motor areas such as the M1, SMA and premotor area that exerts direct control over motor movements. It should also be noted that most TES studies included in this meta-analysis were anodal in nature, and future studies including a cathodal tDCS component will be required to determine any polarity-specific effects of TES on either motor or cognitive function in PD.
While our results demonstrated overall improvements in motor function, several limitations need to be acknowledged. Studies from both rTMS and TES demonstrated modest effect sizes (0.4-0.6) and large heterogeneity between studies. Clinical and lifestyle variables including PD-related comorbidity, physical activity levels and other mental health conditions were not accounted for in our subgroup analyses, which may have influenced the responsiveness to NBS. Other such factors that may have also contributed to heterogeneity between studies is the method of sham stimulation and individual responsiveness to NBS. Most rTMS studies used different coil orientations as a method to provide sham stimulation, however, the validity of these methods has been brought into question given the positive correlation between increasing participant sensation and unwanted cortical activation 77 . Standardisation of a consistent sham condition for rTMS studies is needed to accurately draw conclusions about the efficacy of rTMS improving motor and cognitive symptoms in PD. Furthermore, factors such as age, genetics, circadian rhythm and underlying excitability state are also considerable factors contributing to the variability between studies 78 . Another important factor to consider is that asymmetric disease dominance will undoubtedly influence PD symptomology 79,80 and possibly the responsiveness to NBS. To account for some of these variations, an estimated sample size of more than 30 is needed to detect a reliable between-group difference 78 , however, only 12% of the studies in this review met that criteria and are likely inadequately powered to account for individual differences. Indeed, the figures presenting the results of the meta-analyses show that the confidence intervals for the average effect size is wider for the TES studies compared to rTMS. Additionally, this review included all variant forms of rTMS and TES such as theta-burst stimulation (TBS) and transcranial alternating current stimulation (tACS). Although we consider this a strength by increasing the clinical relevance of our findings, these variant forms of NBS may have modulated motor and cognitive functions through different neural mechanisms. Lastly, to allow for greater homogeneity, our between-study comparisons excluded any follow-up time points in our meta-analyses. Future reviews should aim to determine the longer-term, retention effects of rTMS and TES in PD.
In conclusion, our meta-analysis demonstrated that rTMS and TES are both viable techniques to improve motor symptoms in individuals with PD, particularly gait. Our knowledge towards cognitive improvements from NBS is limited by an insufficient number of studies and inadequate understanding of the neurobiological underpinning of cognitive impairment in PD. There was no differential effect of stimulation parameters on motor or cognitive outcomes in this review. To be able to translate NBS into a viable form of clinical treatment, a better understanding of how different NBS parameters influences motor and cognitive function is necessary to elicit optimal improvements in function in people with PD.

Study design and registration. Our systematic review and meta-analysis adhered to the Preferred
Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (Fig. 1) and is registered with the International Prospective Register of Systematic Reviews (PROSPERO Registration number CRD42016035699).

Literature search strategy.
Studies included in this meta-analysis were searched through PubMed, EMBASE, Web of Science, Google Scholar, Scopus, Library of Congress and Cochrane library from inception to 26 th February 2016, with an additional search updated through to 3 rd February 2017. Keywords used to search the databases include a combination of the following terms "Parkinson's disease", "motor function", "cognitive function", "non-invasive brain stimulation", "repetitive transcranial magnetic stimulation", "transcranial direct current stimulation", "transcranial stimulation", "repetitive TMS", "transcranial DC stimulation", "transcranial electrical stimulation", "rTMS", "tDCS" and "tACS". In addition, the reference lists from existing systematic reviews and meta-analyses, and studies included in these reviews were searched to identify any additional relevant articles.
Inclusion/exclusion criteria for study selection. Figure 5 shows the flow diagram for study inclusion/ exclusion in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. Articles identified from the search were exported into reference management software (EndNote X7; Thompson Scientific, New York, NY, US) and screened for duplication. Following the removal of duplicates, titles and abstracts were screened to identify relevant studies. The inclusion criteria included: 1. Study population included people with idiopathic PD; 2. Interventions using rTMS or TES; 3. Sham stimulation as a control; 4. Outcome measures for motor function including: a. Clinical motor assessments -including UPDRS-III, H&Y, ADL, AIMS, CDRS, PPT, TUG and FOG-Q; b. Gait kinematic as measured using a force platform, accelerometers or 3D motion capture; c. Finger and/or wrist movement using custom-made systems such as goniometers and touchpads. 5. Validated neuropsychological tests to assess: a. Learning and memory; b. Executive functioning; c. Visuospatial abilities; d. Psychomotor speed. 6. cross-over or parallel study design; 7. Written in the English language; 8. Published in peer-reviewed scholarly journals.
The exclusion criteria included: 1. No data available to determine effect sizes, 2. No control group, 3. Case reports, 4. Studies using TES or rTMS combined with another therapy.
Studies were excluded if the title or abstract were not relevant to, or did not fit the inclusion criteria. If the title or abstract was unclear, the article was assessed in its entirety. To increase the accuracy of the screening process, the first 30% of titles and abstracts were reviewed by two assessors (WPT and AMG) with the remaining titles/abstracts screened by WPT. Two authors screened all full text articles and discrepancies were resolved via discussion. Details of all studies included in this study may be found in Supplementary Tables 1 and 2. Methodological quality. Two assessors (AMH and LJ) assessed the methodological quality of all studies using the Physiotherapy Evidence Database (PEDro) rating scale 81 , whilst another two assessors (MM and NAU) assessed the risk of bias of all studies using the Cochrane risk of bias assessment 82 . The PEDro scale (rated 1-10) assesses study quality in five domains; group allocation, blinding, attrition, statistical analyses and data variability. The Cochrane risk of bias tool for randomised controlled trials rates trial quality on seven domains: sequence allocation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting & other sources of bias. Both pairs of assessors reviewed each study independently, and any discrepancies were graded by a third assessor (WPT). Details of the PEDro score and Cochrane risk of bias assessment may be found in Supplementary Tables 3 and 4. Data extraction. For all included articles, data extraction involved the retrieval of study characteristics (author, year, sample size and study design), participant demographics (age, sex, medication and disease severity), stimulation protocols (type, intensity, duration and site) and the test used to quantify motor and cognitive function outcome measures (refer to Table 1 for types of motor and cognitive tests used). Measures of mood and depressive scales (i.e. Hamilton depression scale, Beck depression scale and Geriatric depression scale) were not included in the analysis. For each included study, quantitative data for pre-and post-intervention, for both the stimulation and control (sham) conditions were extracted from the results text, tables and figures. To increase reliability, two authors (AMG & WPT) extracted the data from each study. Where studies reported more than one 'post' assessment, the time-point immediately following the last intervention session was extracted and used in the analysis. If studies reported multiple cognitive or motor tests, an overall weighted mean combining the effect size for each test was used to calculate a single effect size for that study. The data extracted varied depending on the reported results. This included pre-post means and standard deviations (SD; or conversion of SEM to SD) for real and sham NBS and the p-value representing the group (sham vs. real) by time (pre-post) interaction between the stimulation conditions. A plot digitising software (PlotDigitizer) was used to extract data from figures when data was not readily available from text or tables.

Statistical analyses.
As systematic influences and random error were predicted to be present between study level effect sizes, a random effects meta-analysis was performed to compare the overall pooled standardised mean differences (SMDs) for motor and cognitive function between rTMS and TES studies. Separate meta-analyses were conducted for each of the 3 categories of motor outcomes (i.e. gait, hand movements and UPDRS III) for both rTMS and TES studies and presented as averaged SMD and 95% confidence interval (CI) values. Cognitive outcomes for rTMS and TES studies were presented as an overall SMD and 95%CI value due to an insufficient number of studies to perform separate meta-analyses.
For each study, SMD was computed such that positive values indicate that the treatment group's performance on motor and cognitive tests was superior to sham on the outcome variable 83 . Subgroup analyses and meta-regression were agreed upon a-priori to assess the influence of moderator variables of rTMS on cognitive and motor function and moderator variables of TES on motor function. Subgroup and meta-regression analyses were excluded if there were insufficient studies for comparison. Where studies had more than one subgroup comparison, they were removed from the subgroup analysis 84 . These moderator variables included: Heterogeneity was measured using the I 2 statistic, which indicates the percentage variance between studies with cut off points corresponding to low (25%), moderate (50%) and high (75%) heterogeneity 85 . Funnel plots assessed publication bias using Egger's regression test (where non-significant asymmetry indicated no bias) 86 . All statistical analyses were performed using Comprehensive Meta-Analysis (V3.0, Biostat, Englewood, USA) using an alpha level of p < 0.05 was used to determine significance.