Transcranial direct current stimulation for Parkinson’s disease: systematic review and meta-analysis of motor and cognitive effects

Abstract

Transcranial direct current stimulation (tDCS) is a promising noninvasive intervention for Parkinson’s disease (PD). However, studies of its motor and cognitive effect have produced mixed results. We conducted a systematic review including 38 studies and meta-analysis of 12 randomized sham-controlled trials with 263 PD patients. No significant differences were found between active and sham tDCS in motor function (UPDRS-III: SMD = –0.14, p = 0.74), gait (SMD = 0.10, p = 0.513), attention and working memory (SMD = 0.24, p = 0.13), executive function (SMD = 0.03, p = 0.854), and memory and learning (SMD: −0.07, p = 0.758). The prediction intervals indicated substantial heterogeneity among studies. Meta-regression showed small positive effects in younger PD patients with milder symptoms. These findings are preliminary but suggest tDCS may benefit some PD patients while being neutral or harmful to others.

Introduction

Parkinson’s disease (PD) is a progressive brain disease characterized by motor and non-motor symptoms^1–3. In addition to motor symptoms such as tremor, bradykinesia, and rigidity, non-motor symptoms such as autonomic dysfunction and cognitive deficits significantly affect a patient’s quality of life and even prognosis, in particular the cases of dementia^4,5. Similar to motor symptoms, cognitive non-motor symptoms vary individually, ranging from mild cognitive impairment to dementia, and may potentially be predicted by genetic or clinical features⁶. Although dopaminergic medications are available that can help to alleviate motor symptoms in the initial stages of PD, these medications become less effective with illness progression and carry risk for serious adverse effects, such as motor complications and hallucinations⁷. Furthermore, no interventions are available that can improve the cognitive symptoms of PD or halt further cognitive decline. As a result, extensive research is underway to explore new or more advanced therapeutic interventions for PD⁸.

One promising intervention is noninvasive brain stimulation, which includes transcranial magnetic stimulation (TMS) and transcranial electrical stimulation (tES; encompassing both direct current, tDCS, and alternating current, tACS)⁹. TMS, first introduced by Barker et al. in 1985, employs pulsed magnetic fields to stimulate specific cortical regions, thereby modulating neural activity and influencing brain function¹⁰. The evolution of this technology led to the development of repetitive TMS (rTMS), which allows for extended treatment sessions and has significantly enhanced the therapeutic potential of TMS in modulating brain activity¹¹. With tES, patients receive weak (1–2 mA) electric currents aimed at alleviating clinical symptoms by directly modulating brain activity. Although its precise mechanism of action remains elusive, pre and clinical improvements following tES have been attributed to effects on cell membrane potentials, neurotrophic signaling, and long-term potentiation/depression-type neuroplasticity¹². At a macroscopic level, the clinical benefits of tDCS have been linked to its effects on cortical excitability¹³, while tACS has been associated with modulating brain oscillatory dynamics¹⁴, and both improving large-scale motor and cognitive network functioning via corticostriatal circuit¹⁵. Compared to TMS, tES is relatively inexpensive, painless, easy to administer, and portable⁹.

Over the past 15 years, several randomized clinical trials (RCTs) and meta-analytic reviews have been published on motor and cognitive outcomes in patients with PD following tDCS delivery. However, the results of different meta-analytic reviews have been mixed, probably in part due to use of different study inclusion criteria and data analytic approach. For example, some meta-analytic reviews^16,17 report that tDCS given to PD patients produced a significant improvement in overall motor function, as indexed by a reduced score on the Unified Parkinson’s Disease Ranking Scale (UPDRS)-III, but other meta-analyzes report no significant differences^18,19. On the other hand, meta-analytical reviews of cognitive outcomes of tDCS seem to have yielded somewhat more consistent results, including a small but significant improvement on tests of working memory and attention, which was evident in not only PD patients but also patients with other neurological or psychiatric disorders²⁰.

It appears, thus, that tDCS may exert beneficial effects on cognitive functioning in patients with PD while having no or only minor effects on motor symptoms¹⁹. However, such a conclusion would be based primarily on point effect size estimates and corresponding confidence intervals (CIs) and p-values, which do not provide the range of true tDCS effects expected to be observed in similar studies and patients, such as in a next study or in a comparable real-life setting^21,22. In contrast, the prediction interval (PI) conveys this information. Unfortunately, the PI has not yet been utilized to capture variation in the true effect size across different populations of studies and patients included in the meta-analysis. Thus, the PI could disclose large between-study heterogeneity of tDCS intervention effects, which may range from beneficial effects to zero effect in some populations and, perhaps, even effects in the opposite direction of the summary point effect size estimate (i.e., detrimental effects) in other populations²².

In this context, we conducted a systematic review and meta-analysis of studies that utilized tDCS intervention for symptom relief in patients with PD. Our aim was to provide a better understanding of the strength and heterogeneity of the effects of tDCS on patients’ motor and cognitive functioning, as measured by established clinical symptom rating scales or standardized motor and cognitive tests.

Results

From the databases, we identified 1479 records (operators in Supplementary). After removing 749 duplicates, we retained 730 records for screening and selection (Fig. 1). Thirty-eight studies examined tDCS effects in a total of 754 patients with PD (Table 1). Thirty-four of these studies were RCTs and 4 were non-RCTs, including observational studies.

Fig. 1.

Flow diagram depicting search, identification, exclusion, and inclusion of studies for review and meta-analysis.

Table 1.

Studies on motor and cognitive outcomes after tDCS in patients with PD

Reference (author, reference number)	Patient characteristics[age (years), female : male ratio, disease severity, medication state)]	Patient sample size	Study design	Intervention	Target	Motor and cognitive outcome measure	Key reported findings
Boggio et al., 2006a	61.1 ± 11.1, 12:6, HY 2.35 ± 0.8, “off” state	18	RCT crossover Exp 1: n = 9 Exp 2: n = 9	Exp 1:Single, 1 mA, 20 min, anodal Exp 2: Single, 1 mA, 20 min, anodal	M1 or L DLPFC	WM assesment	M1 1 mA: ± 2 mA: +	L DLFPC 1 mA: ± 2 mA: +
Fregni et al., 2006	62.3 ± 1.6, 11:6, HY 2.4 ± 0.2, “on” state	17	RCT crossover	Single, 1 mA, 20 min, anodal and cathodal	M1	UPDRS III, SRT, PPT	Anodal UPDRS III: + SRT: + PPT: ±	Cathodal UPDRS III: ± SRT: ± PPT: ±
Benninger et al., 2010a	63.9 ± 8.9, 9:16, HY 2–4,”on” and “off” state	25	RCT between subjects tDCS n = 13 Sham n = 12	Four sessions in 2·5 w, 20 min, 2 mA, anodal	M1 and PFC	Timed tests of gait and hand and arm movement, UPDRSbradykinesia, UPDRS III, SeRT	“on” Gait time: ± (1d), ± (1 m), ± (3 m) Hand and arm time: + (1d), + (1 m), + (3 m) UPDRSbradykinesia: ± (1d), ± (1 m), ± (3 m) UPDRS III: ± (1d), ± (1 m), ± (3 m) SeRT: ± (1d), ± (1 m), ± (3 m)	“off” Gait time: + (1d), ± (1 m), ± (3 m) Hand and arm time: + (1d), + (1 m), + (3 m) UPDRSbradykinesia: + (1d), ± (1 m), ± (3 m) UPDRS III: ± (1d), ± (1 m), ± (3 m) SeRT: ± (1d), ± (1 m), ± (3 m)
Pereira et al., 2013	61·5 ± 9·9, 9:7, HY 1·6 ± 0·5, MMSE 27·7 ± 2·1, “on” state	16	RCT crossover	Single, 20 min, 2 mA, anodal	L-DLPFC or TPC	PVF, SVF	L DLPFC PVF: ± SVF: +	TPC PVF: ± SVF: +
Valentino et al., 2014	66–76, 1:1, HY 2–4, “on” state	10	RCT crossover	1/d for 5 d, 20 min, 2 mA, anodal	M1	SWS, UPDRS-III, FOG-Q, GFQ	SWS: + (1d) + (5d) + (7d) + (2w) + (4w) UPDRS III: ± (1d) + (5d) + (7d) + (2w) + (4w) FOG-Q: + (7d) + (2w) + (4w) GFQ: + (7d) + (2w) + (4w)
Manenti et al., 2014	67·1 ± 7·2, 4:6, HY1·3 ± 1·1, MMSE 28·5 ± 1·8, “on” state	10	RCT crossover	Single, 7 min, 2 mA, anodal	R DLPFC or L DLPFC	TUG	L DLPFC TUG: ±	R DLPFC TUG: +
Doruk et al., 2014	61 ± 8, 6:12, “on” state	18	RCT between subjects tDCS R DLPFC n = 5 tDCS L DLPFC n = 6 Sham n = 7	10 sessions over 2 w, 20 min, 2 mA, anodal	R DLPFC or L DLPFC	Cognitive tests in EF, VS, WM, UPDRS-III, SRT, CRT, PPT, FT, walking time	L DLPFC EF: ± (immediate) + (3 m) VS, WM, UPDRS-III, SRT, CRT, PPT, FT, walking time: ± (immediate, 3 m)	R DLPFC EF: ± (immediate) + (3 m) VS, WM, UPDRS-III, SRT, CRT, PPT, FT, walking time: ± (immediate, 3 m)
Salimpour et al., 2015	49–75, 6:9, HY1-3, “on” state	10	RCT crossover	Single, 2 mA, 25 min, anodal or cathodal	M1	UPDRS III	Anodal UPDRS III: ±	Cathodal UPDRS III: +
Ferrucci et al., 2016a	74.3 ± 8.0, 4:5, HY 2–3, MMSE 26-30, levodopa induce dyskinesia, “on” state	9	RCT crossover	1/d for 5 d, 20 min, 2 mA, anodal	M1 or cerebellum	UPDRS III, word recall, VAT, SeRT	M1 UPDRS III, word recall, VAT, SeRT: ±	Cerebellum UPDRS III, word recall, VAT, SeRT: ±
Swank et al., 2016a	40–80, 2:8, “on” state	10	RCT crossover	Single, 20 min, 2 mA, anodal	L DLPFC	TUG	TUG: ±
Lattari et al., 2017	69·18 ± 9·98, 4:13, HY 2·35 ± 1·06, “on” state	17	RCT crossover	Single, 20 min, 2 mA, anodal	L DLPFC	BBS, Dynamic gait index and TUG	BBS, Dynamic gait index and TUG: +
Benussi et al., 2017	C: 62·9 ± 10·5 A: 63·2 ± 9·2 S: 63·5 ± 8·6, 17:43, HC: 1·7 + 0·4, A: 1.6 ± 0.5, S: 1·9 ± 0·7, “on” state	60	RCT between subjects tDCS anodal n = 20 tDCS cathodal n = 20 Sham n = 20	Single, 10 min, 2 mA, cathodal or anodal	R DLPFC	IGT	Anodal IGT: ±	Cathodal IGT: +
Cosentino et al., 2017	58 ± 11·5, 1:1, HY 1–2·5, “on” state	14	RCT crossover	Single, 20 min, 2 mA, anodal or cathodal	R or L M1	FT, upper limb bradykinesia, and UPDRS III	Anodal FT: + (L) ± (R) Upperlimb bradykinesia: ± (L) + (R) UPDRS III: + (L) ± (R)	Cathodal FT: ± (L) + (R) Upperlimb bradykinesia: ± (L) + (R) UPDRS III: ± (L) ± (R)
Elder et al., 2017	66·6 ± 8·4, 11:27, “on” state	38	RCT crossover	Single, 20 min, 2·8 mA, anodal	L DLPFC	SRT, CRT, DV, ANT-EF	SRT, CRT, DV, ANT-EF: ±
Hadoush et al., 2018	43–76, 5:13, HY 1–4, “‘on” state	18	observation study	5 sessions/ w 2 w, Bilateral, 20 min, 1 mA, anodal	M1 and DLPFC	BBS, FES-I, 10 m walking test	BBS, FES-I: + 10 m walking test: ±
Dagan et al., 2018	68·8 ± 6·8, 3:17, HY 2–3·5, MMSE ≥ 21, “on” state	20	RCT crossover	Single, 20 min, 1·5 mA, anodal	M1 and DLPFC or M1	FOG-provoking test, TUG, Stroop test	M1 and DLPFC FOG-provoking test: + TUG: + Stroop test: +	M1 FOG-provoking test: ± TUG: ± Stroop test: ±
Hadoush et al., 2018	43–76, 5:15, HY1–5, “on” state	20	Observation study	5 times/w for 2 w, 20 min, 1 mA, bilateral or anodal	M1 and DLPFC	UPDRS-III (total, bradykinesia, tremor, rigidity, and gait/balance/speech sub-scores)	UPDRS-III (total, bradykinesia, tremor, rigidity, and gait/balance/speech sub-scores): +
Da Silva et al., 2018a	50–80, 7:10, HY 2–3, MMES < 18, “on” state	17	RCT between subjects tDCS n = 8 Sham n = 9	Single, 15 min, 2 mA, anodal	M1 and SMA	UPDRS III	UPDRS III: ±
Adenzato et al., 2019	65·6 ± 8·4, 1:1, HY 1·8 ± 0·7, PD with MCI, “on” state	20	RCT crossover	Single, 6 min, 1·5 mA, anodal	MFC	RME, AI	RME, AI: +
Schoellmann et al., 2019	46–80, 4:6, “off” state	10	RCT crossover	Single 20 min, 1 mA, anodal	L sensorimotor area	Segmental UPDRS-III (R hand)	Segmental UPDRS-III (R hand): +
Bueno et al., 2019a	64.5 ± 9.0, 8:12, HY 2.25 ± 0.63, MMSE 27.05 ± 2.83, “on” state	20	RCT crossover	Single, 20 min, 2 mA, anodal	L DLPFC	TMT A and B, SVF, Stroop test, and TUG	TMT A and B, SVF, Stroop test, and TUG: ±
Lau et al., 2019a	56–78, 1:1, HY 2·15 ± 0·3, MMSE 26–27, “on” state	10	RCT between subjects tDCS n = 5 Sham n = 5	Single, 20 min, 2 mA, anodal	L DLPFC	Visual working memory, Go/no-go test	Visual working memory, Go/no-go test: ±
Workman et al., 2020	72·4 ± 6·4, 2:5, HY 1·9 ± 0·4, “on” state	7	RCT crossover	Single, 20 min, unilateral or bilateral, 2 mA or 4 mA, anodal	Cerebellum	BBS	Unilateral BBS: ± (2 mA) ± (4 mA)	Bilateral BBS: ± (2 mA) + (4 mA)
de Albuquerque et al., 2020	71·3 ± 8·6, 12:10, “on” state	22	RCT between subjects tDCS n = 11 Sham n = 11	Single, 25 min, 2 mA, anodal	Cerebellum	UPDRS III	UPDRS III: ±
Firouzi et al., 2021	PD with MCI 70–84, HY II-III, 3:8, “on” state	11	single-blind, sham-controlled crossover	Single, 20 min, 2 mA, anodal	M1	SRT	SRT: ± (5 min) + (1w)
Mishra et al., 2021	49–77, 6;14, HY1–3, “on” state	20	RCT crossover	Single, 30 min, 2 mA, anodal	L DLPFC	PVF, PVFwalking	PVF: ± (during) ± (immediately) + (15 min) ± (30 min) PVFwalking: ± (during) ± (immediately) + (15 min) + (30 min)
Manor et al., 2021a	50–90, 14:59, HY 1–3, “on” and “off” (n = 45) state	73	RCT between subjects tDCS n = 37 Sham n = 36	10 sessions over 2 w Followed by: 1 sessions/w for 5 w, 20 min, 1·5 mA, anodal	L DLPFC and M1	FOG-provoking test, UPDRS-III, TUG, EF, self-reported FOG severity, daily living step counts	FOG-provoking test, UPDRS III, TUG, EF: ± Self-reported FOG severity, daily living step counts: +
Beretta et al., 2021	68·9 ± 8·5, 10: 14, HY ≤ 3, MMSE 27.5, ”on” state	24	RCT crossover	Single, 2 mA, 20 min, anodal	M1	Recovery time to stable position	Recovery time to stable position: +
Ruggiero et al., 2022a	42-77, 4:5, HY 2–3, MMSE 2–30, “on” state	9	RCT between subjects tDCS n = 5 Sham n = 4	1/d for 5 d, 20 min, 2 mA, anodal	Cerebellum	SRT	SRT: ±
de Albuquerque et al., 2022	71·2 ± 8·9, 10:11, HY 2·1 ± 0·7, “on” state	21	RCT between subjects tDCS n = 11 Sham n = 10	9 sections over 2 w, 25 min, 2 mA, anodal	Cerebellum	UPDRS III	UPDRS III: ±
Terenzi et al., 2022	PD with ICD: 70·8 ± 6·5, 1·8 ± 0·6, “on” state PD without ICD: 72·7 (5·2), 1·8 (0·4), “on” state	28	Crossover (DLPFC, M1 and sham) tDCS PD + ICD n = 15 tDCS PD n = 13 tDCS HC n = 15	Single, 20 min, 1·5 mA, anodal	L DLPFC or M1	Reward-craving test, temporal discounting tasks	With ICD Reward-craving test, temporal discounting tasks: ± (M1) ± (L DLPFC)	Without ICD Reward-craving test, temporal discounting tasks: ± (M1) ± (L DLPFC)
Wong et al., 2022a	47–70, 17:19, HY1–3, MMSE ≥ 24, “on” state	36	RCT between subjects tDCS M1 n = 9 tDCS DLPFC n = 9 tDCS cerebellum n = 9 tDCS sham n = 9	Single, 2 mA, 20 minutes, anodal	M1 or DLPFC or cerebellum	TUG	M1 TUG: ±	DLPFC TUG: ±	Cerebellum TUG: ±
Aksu et al., 2022a	65·5 ± 7·5, 9:17, HY 2·25 ± 1, PD-MCI, “on”state.	26	RCT between subjects tDCS n = 13 Sham n = 13	10 sessions for 5 d, 2 mA, 20 min, anodal	L DLPFC	Neuropsychological battery including A, EF, VS, M, and language domains	EF, M (delay recall): + A, VS, M (immediate recall), language: ±
Mishra et al., 2022	67·8 ± 8·3, 6:14, HY 1.9 ± 0.9, “on” state	20	RCT crossover	Single, 2 mA, 30 min, anodal	DLPFC	PVFwalking	PVFwalking: + (during) + (immediately) + (15 min) ± (30 min)
Sadler et al., 2022	63·5 ± 7·2, 6:7, “on” state	11	RCT crossover	Single, 1·5 mA, 10 min, anodal	M1 or SMA	Bradykinesia score	M1 Bradykinesia score: ±	SMA Bradykinesia score: ±
Zhang et al., 2023	67·5 ± 4·9, 7:6, HY 1.8 ± 0·8, “off” state	13	RCT crossover	Single, 1.5 mA, 20 min, anodal, cathodal and bilateral	M1	UPDRS III, PPT, FTMTRS	Anodal UPDRS III: + PPT: ± FTMTRS: ±	Cathodal UPDRS III: ± PPT: + FTMTRS: ±	Bilateral UPDRS III: ± PPT: + FTMTRS: +
de Albuquerque et al., 2023	68·4 ± 11·8, 5:3, HY 2, MoCA 28·31 ± 1·70, “off” state	16	RCT crossover	Single, 2 mA, 25 min, anodal	Cerebellum	UPDRS III	UPDRS III: ±
Simonetta et al., 2024	52·3 ± 4·2, 4:6, HY 1·7 ± 0.6, “on” and “off” state	10	RCT crossover	10 sections, 2 mA, 20 min, anodal	M1	UPDRS III, PD-CRS	UPDRS III: ± PD-CRS: ±

^aIncluded for meta-analysis.

A attention, AI attribution of intentions, ANT attention network task, BBS Berg balance scale, CRT choice reaction time, d day(s), DLPFC dorsolateral prefrontal cortex, DV digit vigilance, EF executive function, FER facial emotion recognition task, FES-I falls efficacy scale-international, FT finger tapping, FTMTRS Fahn–Tolosa–Marin Tremor Rating Scale, HC healthy control, HY Hoehn and Yahr Scale, IGT Iowa gambling task, L left, M memory, M1 motor cortex, MFC medial frontal cortex, min minute(s), MMSE mini mental state examination, PD Parkinson’s disease, PPT Purdue pegboard test, PVF phonemic verbal fluency, R right, RME reading the mind in the eyes, SeRT serial reaction time, SMA supplementary motor cortex, SRT simple reaction time, SVF semantic verbal fluency, SWS stand aalk sit, tDCS transcranial direct current stimulation, TPC temporoparietal cortex, TUG time up and go, UPDRS-III Unified Parkinson’s disease rating scale - III, VAT visual attention task, VAS visual analog scale, VS visuospatial ability, VSC visual shape counting, w week(s), WM working memory. The key findings were expressed by improvement “+”, no statistically significant change “±”, and negative impact “−”.

Twelve of the 38 studies were included in the meta-analysis (7 using a parallel group and 5 a cross-over design), encompassing a total of 263 patients with PD [mean age: 65 years, range: 47–74; mean illness duration: 8.3 years, range: 0.5 to 10 years; mean illness severity/disease stage measured by the Hoehn and Yahr (H-Y) scale: mean 2.2, range: 1.5–2.5]. One study measured working memory in the “off” medication status, two studies regarding motor function examined both “on” and “off” medication status, and the rest were performed with patients “on” medication (Table 1). Motor outcome measures in the 12 RCTs mainly included overall motor function as assessed by the UPDRS-III (n = 5), and gait and balance (n = 5). Cognitive outcome measures primarily fell into the domains of attention and working memory (n = 6), executive function (n = 4), and memory and learning (n = 2). Outcomes were measured immediately after one or more sessions of tDCS, with some studies also including follow-up assessments a few weeks or months later (Table 1).

Quality of included studies

Most studies showed a low risk of bias in the randomization process (D1) and deviations from intended interventions (D2). There was more variability in the risk of bias due to missing outcome data (D3), with several studies showing some concerns or high risk. The measurement of outcomes (D4) was mostly at low risk, while the selection of reported results (D5) presented some concerns across multiple studies.

The highest levels of bias were observed in the domains of missing outcome data and measurement of outcomes, with several studies showing high or some concerns in these areas. The details of the risk of bias of all the included studies were available in Fig. 2a and summarized in Fig. 2b.

Fig. 2. Risk of bias summary.

a “Traffic light” plots of the domain-level judgements for each individual result. b Weighted bar plots of the distribution of risk-of-bias judgements within each bias domain.

Motor Outcomes

Figure 3 presents forest plots illustrating the individual and weighted average effect size estimates for the outcome measures of overall motor function (UPDRS-III score) (panel a) and gait and balance (panel b).

Fig. 3. Forest plots of individual and weighted average effect size estimates (Hedge’s g).

a Outcome measures of overall motor function (UPDRS-III score). b Outcome measures of gait and balance.

UPDRS-III score

For the UPDRS-III score, Hedge’s g pooled across 5 studies [77 patients after active tDCS, 76 patients after sham tDCS; target: premotor-M1 area + DLPFC (k = 1), M1 + DLPFC (k = 1), M1 + SMA (k = 1), M1 alone and cerebellum alone (k = 1), M1 alone (k = 1)] reached a value of −0.14 (95% CI: −1.23 to 0.95) (Fig. 3a). This result indicates that the mean UPDRS-III score tended to be slightly higher (toward motor impairment) after active tDCS compared to sham tDCS, but the mean difference was not statistically significant (t = −0.36, df = 4, p = 0.737). However, considerable effect size heterogeneity was evident (tau-squared = 0.64, I-squared = 0.84), along with a significant Q-test statistic (chi-square = 25.93, df = 4, p < 0.001), rejecting the null hypothesis of homogeneous true effect sizes across studies. Similarly, the 95% PI (−2.97 to 2.69) was substantially wider than the 95% CI. The I-squared index signified that up to 84% of the total variance in observed tDCS effects on the UPDRS-III score reflected variance in true effects across studies rather than random sampling error.

A sensitivity analysis indicated that the mean effect estimate reached values of −0.48 (SE = 0.28, k = 4) and 0.10 (SE = 0.42, k = 4) when excluding the study displaying the largest positive effect (g = 1.24, weight = 19%; Simonette et al.) and the study with the largest negative effect (g = −0.99, weight = 22%; Manor et al.), respectively. However, heterogeneity was still marked after excluding either of these studies (tau-squared = 0.20 and 0.58, respectively).

Consequently, random-effects meta-regression was used to identify sources of effect size heterogeneity. Five potential moderators were considered: age, illness duration, illness stage, motor symptom severity at baseline, and the number of tDCS sessions. Two variables emerged as effect-modifiers: patients’ severity of motor symptoms and age at baseline (Fig. 4). Higher UPDRS-III scores at baseline were associated with progressively less positive (i.e., more negative) effect sizes (Fig. 4a). Specifically, as patients’ baseline motor symptom severity scores increased by 27 points (from score 13 to 40), the predicted effect size decreased by up to 156 points (1.56 SD units, from 0.53 to −1.03). The meta-regression model with symptom severity at baseline as the single covariate explained more than half of the total between-study effect size heterogeneity (R-squared analogous = 59%), although significant residual heterogeneity remained (tau-squared = 0.26, I-squared = 0.64; chi-squared = 8.22, df = 3, p = 0.042). However, we could not formally reject the null hypothesis that the true value of the slope of the meta-regression prediction line was zero (baseline UPDRS-III score: beta = −0.06, t = −2.20, df = 3, p = 0.116, 95% CI: −0.14 to 0.03).

Fig. 4. Scatterplots illustrating meta-regression prediction lines relating individual motor effect size estimates (Hedge’s g).

a Meta-regression to the baseline UPDRS-III score of patient. b Meta-regression to the age of patient.

Similarly, increasing age at baseline was associated with progressively less positive (i.e., more negative) effect sizes (beta = −0.07, t = −1.46, df = 3, p = 0.240, 95% CI: −0.22 to 0.08, R-squared analogous = 25%). The predicted effect size decreased by a total of 149 points (1.49 SD units, from 0.78 to −0.71) across the study-level age range of 52 to 74 years (Fig. 4b). By contrast, meta-regression models including illness duration (range: 5.5 to 10.8 years, k = 5), illness stage (range: 1.5 to 2.5 on H-Y scale, k = 5), or number of stimulation sessions (values: 1, 5, or 10 sessions, k = 5) did not account for any heterogeneity variance (all three R-squared analogous values = 0%).

Gait and balance

For gait and balance, effect size estimates were obtained from 5 studies [79 patients 1after active tDCS, 77 patients after sham tDCS; target: premotor-M1 area + DLPFC (k = 1), M1 + DLPFC (k = 1), DLPFC (k = 2), M1 alone, DLPFC alone, and cerebellum alone (k = 1)]. The mean intervention effect reached a value of 0.10 (t = 0.72, df = 4, p = 0.513), indicating a slight, but not significant, improvement in patients’ timed TUG performance after active tDCS compared to sham tDCS (Fig. 3b). The 95% CI (−0.28 to 0.48) was wide, reflecting low precision of the mean effect estimate, and heterogeneity appeared trivial (tau- and I-squared = 0).

A sensitivity analysis indicated that the observed mean effect (g = 0.10, SE = 0.14, k = 5) showed modest changes when excluding the study with the largest positive effect (g = 0.35, weight = 9%; Wong et al., g = 0.07, SE = 0.14, k = 4) and the study with the largest negative effect (g = −0.02, weight = 22%; Bueno et al., g = 0.13, SE = 0.16, k = 4). The effect size estimate from Wong et al. was also characterized by poor precision, suggesting possible reporting bias.

Inspection of a funnel plot under the fixed-effect model suggested that reporting bias might have inflated the observed mean intervention effect, as effect estimates from studies with similar precision were not symmetrically distributed around the mean effect size (Fig. 5a). Assuming neither heterogeneity nor chance caused the apparent funnel plot asymmetry, some studies with relatively low precision seemed to be missing above the point of zero effect and in areas of small negative effects. A trim-and-fill analysis estimated the number of missing studies to be 2, producing an adjusted fixed-effect point estimate of 0.05 (k = 7, 5 observed and 2 imputed studies, z = 0.38, p = 0.706, 95% CI: −0.20 to 0.29). Thus, the small positive mean effect size initially observed for gait measures (g = 0.10, k = 5) likely overestimates the real effect of active tDCS intervention.

Fig. 5. Funnel plot of individual effect size estimates (Hedge’s g) plotted against standard errors.

a Global motor score. b Global cognitive score.

Other studies of motor outcomes

Eight studies of tDCS for PD not included in the meta-analysis also utilized the UPDRS-III score as an outcome measure (Table 1). Four out of the 8 studies (50%) found significant reductions in the UPDRS-III after tDCS, mainly using M1 as the target of stimulation^23–26. The other four studies (50%) did not find significant effects after tDCS applied to DLPFC²⁷ or cerebellum^28–30.

Seven studies examined gait and balance after tDCS (Table 1). Two out of these 7 studies (29%) found significant improvements in patients’ gait and balance scores after stimulating M1^24,31. Three studies (43%) observed improvement only with specific settings or durations targeting M1²⁶, cerebellum³², or DLPFC³³. Two studies (29%) did not report significant effects after stimulating M1²³ or DLPFC²⁷.

Finally, three studies examined upper limb function in PD patients after tDCS to various targets: M1 (n = 1), SMA (n = 1), sensorimotor cortex (n = 1), and DLPFC (n = 1) (Table 1). One study (33%) found significant improvements after stimulating the sensorimotor cortex³⁴, whereas the other two studies (66%) detected no significant differences following tDCS after M1 or SMA^27,35.

Cognitive Outcomes

Figure 6 presents forest plots illustrating the individual and weighted average effect size estimates for measures of attention and working memory (panel a), executive function (panel b).

Fig. 6. Funnel plot of individual effect and weighted average effect size estimates (Hedge’s g).

a Outcome measures of attention and working memory. b Outcome measures of executive function.

Attention and working memory

Hedges’ g for measures of attention and working memory, pooled across 7 studies [69 patients after active tDCS, 67 patients after sham tDCS; target: DLPFC (k = 3), premotor-M1 area + DLPFC (k = 1), M1 alone, DLPFC alone (k = 2), cerebellum (k = 1)], reached a value of 0.24 (t = 1.77, p = 0.128, 95% CI: −0.09 to 0.58) (Fig. 6a). This finding indicates that the patients’ level of attention/working memory performance after active tDCS was higher, though not significantly, than the performance level after sham tDCS. No evidence was found that the true effect size varied across studies (tau- and I-squared = 0).

A sensitivity analysis showed that, relative to the initially observed mean effect size (g = 0.24, k = 7), the mean effect estimate reached a moderately lower value (g = 0.18, 95% CI: −0.21 to 0.57, k = 6) and a higher value (g = 0.30, 95% CI: −0.08 to 0.67, k = 6) when excluding the studies with the highest positive effect size (Hedges’ g = 0.48, weight = 15%; Boggio et al.) and the largest negative effect size (Hedges’ g = −0.19, weight = 20%; Lau et al.), respectively. Despite the wide range of observed effect sizes, a funnel plot under the fixed-effect model demonstrated that all studies were located within the expected triangular region, suggesting no heterogeneity or reporting bias (Fig. 5b). The individual effect size estimates were symmetrically scattered around the mean fixed-effect size (g = 0.24), further indicating the absence of heterogeneity and reporting bias.

Executive function

For measures of executive function, Hedges’ g pooled across 4 studies [43 patients active tDCS, 43 patients sham tDCS; target: DLPFC (k = 4)] reached a nonsignificant value of 0.03 (t = 0.20, df = 3, p = 0.854) (Fig. 6b). The margin of error for the estimated mean effect size was large, with the 95% CI (−0.50 to 0.57) encompassing the null effect as well as small-to-medium negative and positive effects. No evidence of between-study effect size heterogeneity (tau- and I-squared = 0) or reporting bias was found, as the effect estimates visualized in a funnel plot were symmetrically distributed around the mean effect size.

Memory and Learning

There were only 2 studies available for estimating the intervention effect on measures of memory and learning [22 patients active tDCS, 22 patients sham tDCS; target: DLPFC (k = 1), M1 alone, cerebellum alone (k = 1)]. The weighted average effect size estimate was −0.07 (t = −0.40, df = 1, p = 0.758, 95% CI: −2.27 to 2.13), reflecting a nonsignificant trend toward impaired memory/learning performance following active tDCS. The wide 95% CI indicated poor precision of the point estimate, suggesting that the intervention effect in the population might actually be zero.

Other studies of cognitive outcomes

Four studies not included in the meta-analysis examined attention and working memory in PD patients following tDCS to DLPFC (n = 2) and M1 (n = 2) (Table 1). Two studies (50%) observed improvements by targeting M1 with certain stimulation conditions or at one of the measuring time points but not with others^23,36, while the remaining two studies (50%) did not detect significant differences^27,37.

Seven studies examined executive functioning in PD patients following tDCS to DLPFC (n = 6), M1 (n = 1), TPC (n = 1), and MFC (n = 1) (Table 1). One study (14%) reported significant improvements in tests of executive functioning after tDCS on MFC³⁸. Four studies (57%) observed improvements with certain stimulation conditions, measuring times, or in one of the examination tests but not others after stimulating DLPFC^27,39–41 and TPC³⁹. The remaining two studies (29%) did not detect significant differences after stimulating DLPFC^37,42 or M1⁴².

One study targeted DLPFC and measured visuospatial function but detected no significant improvement after the stimulation²⁷.

Discussion

This meta-analytical review found that patients with PD who received tDCS intervention did not differ significantly from those who received sham tDCS. The meta-regression model disclosed three possible types of tDCS effects: small positive effects (effect size of 0.32) in patients with mild motor symptoms (UPDRS-III score at 13); null effects in patients with moderate symptoms (UPDRS-III scores 22–25); and small-to-medium negative effects (effect size of −0.48) in patients with severe symptoms (UPDRS-III score of 40). A similar tendency was observed with age, showing a near null effect at 64 years, while illness duration, illness stage, or number of stimulation sessions did not account for any heterogeneity variance. The findings suggest tDCS could potentially benefit younger patients with less severe symptoms.

Additionally, results from other qualitative studies, including observational studies, highlighted that tDCS on M1 might improve UPDRS score^23–26 as well as gait and balance^24,31. However, the evidence is not compelling due to large variability and inconsistent effects within and across studies. The inconsistent findings were also reported in 3 studies provided long-term follow up including Benninger et al. reported significant improvement in gait on the first day but no improvement after one month, while Valentino et al. observed consistent improvement up to four weeks. Failure to improve executive function was noted in two studies from immediate post-stimulation to three-month follow-up^27,43.

Assuming that tDCS effects are specific to individual patients, focusing on inter- and intra-individual differences such as genetic background or comorbid conditions could offer a personalized and potentially more fruitful approach than relying solely on group-averaged data and summary effect size estimates. In particular, genetic polymorphisms (e.g. LRRK2, PARK2, SNCA, DJ-1, COMT and ALDH2) or socio-cultural factors that vary across individuals or ethnic groups may not only contribute to differences in symptom manifestation and drug efficacy but could also influence the therapeutic outcomes of neuromodulation^44,45. To examine this possibility in more detail, it would be useful if future studies routinely analyze and present individual patient data. Moreover, extensive research, particularly large-scale multicenter trials with patient characteristic-based subgroup analyzes, is needed to identify and understand sources of between-study variability in tDCS effects, including clinical and methodological differences such as patient sample characteristics, tDCS parameters and protocols, duration of follow-up, and outcome measurement.

It seems plausible to assume that the large variability in tDCS effects across studies and patients originates partly from the manner in which the electric current stimulation has been delivered to patients, posing significant technological and methodological challenges. In this context, it may be informative to contrast tDCS with deep brain stimulation (DBS) targeting the globus pallidus interna or subthalamic nucleus, which has been successfully applied to the management of severely disabled PD patients with medication-refractory motor symptoms and complications⁴⁶. First, DBS is a neurosurgical implant that usually delivers continuous and sustained electric stimulation over years, whereas tDCS is most commonly given only in a number of sessions occurring over days or weeks. Second, DBS stimulation parameters, such as intensity, are carefully adjusted for individual patients during and after surgery to improve clinical benefits while minimizing side effects, whereas tDCS parameters are typically fixed across the period of intervention and not personalized. Third, DBS is surgically implanted into a precisely targeted brain region using detailed anatomical information obtained from individual MR images. In contrast, the spatial resolution of tDCS is severely limited by volume-conduction effects, which shunt, weaken, and distort scalp-applied electric currents flowing from the stimulation electrode into the skin, subcutaneous tissue, skull, cerebrospinal fluid, and brain^47,48. Additionally, tDCS is typically delivered via relatively large stimulation electrodes, which may not always be positioned exactly across participants and studies⁴⁹, inducing cortical activation patterns likely to extend well beyond the focal area of interest⁵⁰. Finally, the possibility should be considered that tDCS in its current form does not evoke any direct brain changes in human participants, which is not an issue of significant concern in DBS. On this view, the transient motor and cognitive effects seen in human participants following tDCS could originate from indirect and nonspecific mechanisms, including placebo effects⁵¹, differences in task strategy (e.g., speed-accuracy tradeoff), effects on comorbid mood symptoms, peripheral effects⁵², and heightened arousal and alertness⁵³.

Even with DBS showing promise, improving cognitive function remains challenging and sometimes even has negative impacts⁵⁴. Cognitive functions are difficult to modulate with unspecific stimulation focusing on a single cortical target. In Parkinson’s disease, degeneration of multiple neurotransmitter systems, including dopaminergic, cholinergic, and noradrenergic systems, plays a critical role in cognitive impairment⁵⁵. Loss of the lateral dopaminergic system in frontal, parietal and temporal cortical regions were found in PD patient with cognitive decline⁵⁶. The extended interplay of these neurotransmitter systems as well as their networking effects make the cognitive function difficult to handle. Considering this, more targeted and precise brain stimulation techniques are needed to effectively address cognitive decline.

Yet, the assessment and reporting of adverse events following tDCS have not always received adequate consideration in tDCS studies⁵⁷, and unknown risks remain. However, to date, only transient and reversible side effects of tDCS have been reported, and not significant or enduring negative effects⁵⁸. Intervention in patients with PD remains unclear since the initial promising findings reported more than 15 years ago^23,43. In fact, based on the present meta-analysis, as well as on prior meta-analyzes, the null hypothesis that tDCS intervention has no genuine effect on patients’ gait and cognitive function can still not be rejected.

The present results, however, should be viewed as preliminary due to the limited number of studies and patients included. We also found some tentative evidence for funnel plot asymmetry, especially for cognitive outcomes where null and negative findings appeared to be underrepresented, indicating the presence of publication bias. This may have inflated effect sizes, although other sources of funnel plot asymmetry, such as chance, cannot be excluded, given the small number of studies included. Furthermore, statistical power of most studies is typically low due to small patient samples, making it possible to detect only large tDCS effects, which may not be realistic to emerge, while the number of tests and intervention concealment and placebo effects remain issues of concern.

In this meta-analytic review, we found insufficient evidence for clinically significant effects of tDCS on motor and cognitive functions in patients with PD. Given the technological and methodological challenges involved in noninvasive electric stimulation therapy, along with an incomplete understanding of volume-conduction effects, further research is required to identify and understand sources of between-study variability in tDCS effects. Novel non-invasive methods like transcranial temporal interference stimulation could improve the accuracy of transcranial electrical stimulation, deserving more focus for developing customized and potent therapies for Parkinson's disease⁵⁹.

Methods

Search strategy

We conducted a comprehensive literature search using the key terms “transcranial alternating current stimulation”, “tACS”, “transcranial direct current stimulation”, “tDCS”, “Parkinson*”, and “PD”. These terms were submitted to the PubMed, Web of Science, Cochrane Library, and EMBASE databases up to April 29, 2024. This approach allowed us to capture diverse geographical and disciplinary perspectives, ensuring the inclusion of high-quality, peer-reviewed content, therefore, enhancing the reliability and validity of our study. Additionally, we reviewed studies from relevant previous meta-analyzes to expand our pool of potential records.

Selection criteria

Two Authors independently screened the titles and abstracts of retrieved records to determine eligibility. Both RCTs and non-RCTs were included if they met the following criteria: (1) involved patients with idiopathic PD in inpatient or outpatient settings; (2) used tDCS intervention; (3) for RCTs, compared active and sham tDCS interventions using either a between-group or cross-over design; and (4) assessed clinical outcomes related to motor or cognitive function using established clinical rating scales (e.g., UPDRS-III) or tests (e.g., digit span).

Exclusion criteria were: studies that did not involve PD patients or included PD patients with physical or psychiatric comorbidities, or received co-interventions such as TMS, physiotherapy, cognitive training, or antipsychotic medication (excluding antiparkinson medications). Studies not published in English, animal research, and irrelevant article types (reviews, meta-analyzes, abstracts, and study protocols) were also excluded. Additionally, studies using physiological measures of motor or cognitive outcomes following tDCS were excluded to maintain a focus on behavioral and clinical outcomes, which are directly relevant to the therapeutic efficacy of tDCS in PD. The exclusion of studies employing physiological measures enhances the homogeneity of the included studies, facilitating a more accurate synthesis of the clinical impact of tDCS. tDCS studies with insufficient quantitative data for effect size calculation underwent qualitative analysis. Finally, tACS studies were excluded due to the low number of studies (n = 1) .

Data extraction

Data extracted from each clinical study comprised: (a) study authors, publication date, study design (between- or within-subjects), patient sample size, and drop-out rate; (b) demographic (age, gender) and clinical characteristics (e.g., symptom severity, medication status, illness duration); (c) tDCS parameters and protocol (e.g., scalp region of interest, anodal/cathodal stimulation, stimulation intensity, number and duration of sessions); (d) instruments used to measure motor or cognitive outcomes (e.g., clinical rating scale, motor or cognitive test); and (e) quantitative outcome data (e.g., means, standard deviations) from active and sham tDCS, and qualitative data expressed as improvement “+”, no statistically significant change “±”, and negative impact “−”.

Including parameters such as symptom severity and medication status is crucial as they significantly influence the response to tDCS and help understand baseline characteristics and potential confounders in clinical outcomes. Symptom severity provides insight into the initial condition of patients, important for evaluating the magnitude of improvement. Medication status is critical since concurrent treatments can modulate the effects of tDCS, impacting the study’s findings. Similarly, specific tDCS parameters (e.g., intensity, duration, and stimulation site) are key to replicating and understanding the heterogeneity in study outcomes. We aimed to classify each cognitive outcome into one of five cognitive domains relevant to PD: attention and working memory, executive function, language, memory and learning, and visuospatial function⁶⁰.

Meta-analysis

Standardized mean difference (SMD) effect size estimates (Hedge’s g with standard error and 95% CI) between active tDCS and sham tDCS groups were computed for each outcome using Comprehensive Meta-Analysis (CMA). We accounted for study design (between- or within-subjects) when calculating Hedge’s g to ensure comparability of effects size estimates from studies using different designs⁶¹. Individual Hedge’s g values were averaged across studies using an inverse variance method and analyzed with a random-effects model in SPSS (version 28.0). The 95% CI indicated the precision of the estimated mean effect size, while the 95% PI (from CMI Prediction Intervals software) measured between-study heterogeneity^21,22. In the absence of between-study heterogeneity, the PI coincides with the respective CI. In the presence of heterogeneity, however, the prediction covers a wider range than the CI^21,22. Cochran’s Q and Higgins’s I2 tests assessed statistical heterogeneity²¹. If heterogeneity was evident, we conducted meta-regression analyzes to identify sources of variability, considering age, symptom severity, illness severity, illness duration, and tDCS parameters (intensity, duration, and sessions) as potential moderators. Sensitivity analysis evaluated each study’s influence by excluding one study at a time. Publication bias was assessed via funnel plots and Egger’s test for funnel plot asymmetry.

Study Quality Assessment

The methodological quality of each study was evaluated using the Cochrane risk of bias tool for individually-randomized clinical trials (RoB-2)⁶² covering 5 domains: (1) risk of bias from the randomization process, (2) risk of bias due to deviations from intended interventions, (3) risk of bias from missing outcome data, (4) risk of bias in outcome measurement, and (5) risk of bias in selecting the reported result. Each domain was rated as ‘low risk’, ‘some concerns’, or ‘high risk’ of bias. Judgments were based on answers to signaling questions addressing systematic errors, independently assessed by two reviewers. Discrepancies were resolved through consensus. All reviewers adhered to the Methodological Expectations for Cochrane Intervention Reviews (MECIR) standards for assessing risk of bias

Author contributions

Z.D. and C.Z. contributed to conceptualization, synthesize, review and agreed with the manuscript.

Data availability

All data generated or analyzed during this study are included in this published article.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41531-024-00821-z.