Noninvasive brain stimulation in children and adults with attention-deficit/hyperactivity disorder: a systematic review and meta-analysis

Abstract

Background

Repetitive transcranial magnetic stimulation (rTMS) or transcranial direct current stimulation (tDCS) could provide treatment alternatives to stimulant medication for attention-deficit/hyperactivity disorder (ADHD), given some evidence for improvements in cognition and clinical symptoms. However, despite a lack of solid evidence for their use, rTMS and tDCS are already offered clinically and commercially in ADHD. This systematic review and meta-analysis aimed to critically appraise rTMS and tDCS studies in ADHD to inform good research and clinical practice.

Methods

A systematic search (up to February 2019) identified 18 studies (rTMS 4, tDCS 14; 311 children and adults with ADHD) stimulating mainly the dorsolateral prefrontal cortex (dlPFC). We included 12 anodal tDCS studies (232 children and adults with ADHD) in 3 random-effects meta-analyses of cognitive measures of attention, inhibition and processing speed.

Results

The review of rTMS and tDCS showed positive effects in some functions but not others, and little evidence for clinical improvement. The meta-analyses of 1 to 5 sessions of anodal tDCS over mainly the left or bilateral dlPFC showed trend-level improvements in inhibition and processing speed, but not in attention.

Limitations

Heterogeneity in stimulation parameters, patient age and outcome measures limited the interpretation of findings.

Conclusion

The review and meta-analysis showed limited evidence that 1 to 5 sessions of rTMS and tDCS, mostly of the dlPFC, improved clinical or cognitive measures of ADHD. These findings did not support using rTMS or tDCS of the dlPFC as an alternative neurotherapy for ADHD as yet. Larger, multi-session stimulation studies identifying more optimal sites and stimulation parameters in combination with cognitive training could achieve larger effects.

Introduction

Attention-deficit/hyperactivity disorder (ADHD) is defined by age-inappropriate and impairing symptoms of inattention, hyperactivity and impulsivity.¹ It is one of the most common childhood disorders, with a prevalence of approximately 7%; problems persist into adulthood in the majority of cases, and ADHD is associated with poor academic and social outcomes.²

Patients with ADHD have cognitive deficits, most prominently in executive functions, such as motor and interference inhibition, selective and sustained attention, working memory and switching, as well as in timing processes and reward-based decision-making.^3,4

Furthermore, meta-analyses of functional MRI (fMRI) studies in ADHD show underactivation in different cognitive-domain-dependent frontostriatal and frontocerebellar systems, such as the right inferior and medial prefrontal and striatal regions during cognitive control;^5,6 the right dorsolateral prefrontal cortex (dlPFC) and striatal and parietal regions during attention;⁶ the bilateral superior prefrontal regions during working memory;⁷ the inferior frontal, parietal and cerebellar regions during timing processes; ⁸ and the ventromedial frontostriatal areas during reward-related functions.⁹

The most effective short-term treatment for ADHD is with psychostimulant medications,¹⁰ but they have side effects and limited longer-term efficacy.¹¹ Alternative treatments, including behavioural therapies, cognitive training, neurofeedback or dietary interventions, have shown limited efficacy.^12–14

Noninvasive brain stimulation treatments, such as repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS), are promising because they can stimulate key brain dysfunctions that have been established in ADHD over the last 2 decades of fMRI research. ⁹ They are also relatively safe, with minimal side effects; they are cheaper than long-term drug treatments or fMRI neurofeedback, for example; and, more importantly, they can induce neuroplasticity,¹⁵ providing hope for longer-term effects, which drugs do not offer.⁹

In rTMS, rapid magnetic pulses are delivered to the scalp with a wire coil to generate an electric current in the brain via electromagnetic induction. The induced electrical current can trigger action potentials in a focal cortical region under the coil, and when pulses are administered at a particular frequency, rTMS can modulate neural activity with longer-lasting after-effects. In general, high-frequency rTMS (5 to 20 Hz) promotes cortical excitability, and low-frequency rTMS (1 Hz) inhibits cortical excitability.¹⁵ Longer-term clinical improvements with rTMS have been demonstrated in several psychiatric disorders: up to 3 months in obsessive–compulsive disorder when stimulating prefrontal, orbitofrontal and supplementary motor regions;¹⁶ 4 months in schizophrenia when stimulating temporoparietal regions;¹⁷ and 12 months in major depressive disorder when stimulating the dlPFC,^18–20 supporting its neuroplastic potential. Relative to tDCS, rTMS has greater specificity in targeting neural regions,²¹ but it is more expensive because of device costs and extensive user-training requirements. ^22–24 The most common adverse effects are transient scalp discomfort underneath the coil as a result of stimulation of the pericranial muscles and peripheral nerves.^25,26

In tDCS, a weak direct electric current is passed between 2 electrodes (a positive anode and a negative cathode) placed on the scalp. The current modulates spontaneous discharge rates and therefore neuronal network activity by causing subthreshold polarity-dependent shifts in resting membrane potentials, with net increases (anodal stimulation) or decreases (cathodal stimulation) in the excitability of underlying neurons, leading to respective increases or decreases in cortical function and synaptic strength.²⁷ Although tDCS is a relatively new form of noninvasive brain stimulation, there is evidence that it can enhance cognitive functions in healthy controls,²⁸ with longer-term effects of up to 9 or 12 months.^29,30 In psychiatric disorders, positive clinical effects have been observed typically up to 1 month after stimulation (for reviews, see Moffa and colleagues,³¹ Tortella and colleagues³² and Kekic and colleagues),³³ although it is possible that effects are longer-term. However, overall, tDCS effects are often small, especially when administered in single sessions in healthy controls.^34,35 Relative to rTMS, tDCS is cheaper, easier to use and produces relatively less discomfort; the most common adverse effects are mild transient tingling, itching and reddening of the skin underneath the electrodes.³⁶

Both rTMS and tDCS potentiate cellular and molecular mechanisms involved in use-dependent local and distant synaptic plasticity (e.g., γ-aminobutyric acid [GABA] and glutamate-mediated long-term potentiation), which may lead to longer-term effects.²⁰ This systematic review and meta-analysis focuses on the clinical and cognitive benefits of rTMS and tDCS, because they are the most investigated noninvasive brain stimulation methods in ADHD. To our knowledge, other methods have not been as well investigated or applied in clinical settings in ADHD (e.g., intermittent or continuous theta burst stimulation, transcranial alternating current stimulation or transcranial random noise stimulation).

There has been an increase over the last decade of 18 rTMS and tDCS studies in ADHD. Studies have used relatively heterogenous stimulation protocols, as well as clinical and cognitive outcome measures, and have reported mixed positive and negative effects on cognition and ADHD symptoms.^9,37 Furthermore, knowledge is lacking with respect to optimal stimulation protocols for children with ADHD (such as stimulation frequency, intensity or stimulation site), and there are neuroethical concerns about potential costs to nontargeted functions. ³⁸ Despite these shortcomings, however, noninvasive brain stimulation is already offered in private clinics in several countries and is available commercially and online.^39,40

A recent meta-analysis of 10 tDCS studies in ADHD (n = 201) found that predominantly anodal tDCS of the dlPFC led to a significant but small improvement in measures of inhibitory control (Hedges’ g = 0.12, 95% confidence interval [CI] 0.01–0.24), and to a significant but moderate improvement in reaction times in n-back tasks in a smaller meta-analysis of 7 effect sizes derived from 3 studies (Hedges’ g = 0.66, 95% CI 0.17–1.25).⁴¹ However, effect size estimates may have been inflated, because 2 studies reporting mostly null effects were not included;^42,43 multiple dependent effects were clustered in the meta-analyses, unduly reducing variation between effect sizes and overestimating statistical significance;^44,45 and the analysis of inhibitory control measures included noninhibitory measures, such as inattention (e.g., omission errors), processing speed (e.g., reaction times to go trials) and reaction time variability, calling into question the specificity of the positive tDCS effects to the domain of inhibitory control.

We therefore considered it paramount and timely to conduct a systematic review of rTMS and tDCS studies and a meta-analysis of tDCS studies in ADHD that included all available empirical studies and controlled for multiple dependent effects and potential bias; that clustered cognitive effects into clearly separated cognitive domains of inhibitory control, attention and processing speed to elucidate effects of tDCS on specific cognitive domains; that tested the replicability of meta-analysis results with jackknife sensitivity analyses; and that included further sensitivity analyses to reduce heterogeneity caused by studies with designs that deviated from the majority.

This review and meta-analysis aimed to provide a critical appraisal of the most consistent clinical and cognitive effects of rTMS and tDCS in ADHD, scrutinizing the quality of the studies, outlining limitations in the field and discussing neuroethical concerns and future directions, with the ultimate aim of guiding clinical practice.

Methods

Eligibility criteria

The systematic review followed Preferred Reporting Items in Systematic Reviews and Meta-Analyses (PRISMA) guidelines⁴⁶ (Fig. 1; PRISMA checklist, Appendix 1, available at jpn. ca/190179-a1). Inclusion criteria were as follows: empirical studies with sufficient method details that applied rTMS or tDCS in children and/or adults with ADHD confirmed by either a clinical diagnosis (as defined by DSM/ICD criteria) or by meeting cut-off criteria for ADHD on validated ADHD rating scales or research diagnosis questionnaires (e.g., Conners’ Adult ADHD Rating Scale t score > 65⁴⁷ or the Kiddie Schedule for Affective Disorders and Schizophrenia⁴⁸). To avoid language bias, we did not exclude studies published in a language other than English, unless the paper could not be accessed and/or translated by the authors of the article. Relevant outcome measures included clinical measures of ADHD and performance measures on cognitive tasks.

Fig. 1.

Preferred Reporting Items in Systematic Reviews and Meta-Analyses (PRISMA) flow diagram of study selection.

Search strategy

We searched Web of Knowledge, Scopus, PubMed, Ovid, Google Scholar, psyarxiv and bioRxiv (up to the end of February 2019) using the following keywords: “noninvasive brain stimulation,” “transcranial electric stimulation,” “transcranial direct current stimulation,” “tDCS,” “transcranial magnetic stimulation,” “rTMS” or “transcranial electric stimulation,” each in combination with “hyperkinetic disorder,” “attention-deficit/hyperactivity disorder,” “ADHD,” “inattention,” “hyperactivity” or “impulsivity.” We also hand-searched the reference lists of retrieved articles and reviews. One author (S.J.W.) and an additional reviewer (S.W.H.) carried out the search separately; another author (K.R.) crosschecked the results.

Study selection

After removing all duplicates, 1 author (S.J.W.) and an additional reviewer (S.W.H.) independently screened titles and abstracts. The full text of the remaining studies determined final inclusion in accordance with our eligibility criteria. Of the 16 778 studies identified, 14 177 duplicates and 2572 irrelevant papers were excluded after screening titles and abstracts. A full-text review of the remaining 29 studies resulted in the exclusion of a further 11 studies, including 1 rTMS study written in Arabic, because the paper could not be accessed and a translation could not be obtained from the authors⁴⁹ (Appendix 1, Table S1). This resulted in 18 peer-reviewed, published studies (4 rTMS and 14 tDCS; total ADHD sample = 312), of which 12 anodal tDCS studies (total n = 232) were eligible for meta-analysis (for the list of excluded studies, see Appendix 1, Table S1). The final list of included studies was agreed upon by consensus; any disagreements were resolved by another author (K.R.).

Risk of bias assessment

We assessed risk of bias using the Cochrane Collaboration’s risk of bias tool,⁵⁰ which rates risk of bias across 5 domains: selection bias, performance bias, detection bias, attrition bias and other biases. Using this tool, S.J.W. and K.R. assessed risk of bias for all sham-controlled studies and resolved any disagreements by consensus. Because the tool is designed for randomized controlled trials, we excluded open-label trials^51,52 to avoid undue inflation of bias across domains. We included open-label trials in the systematic review for a complete overview of available empirical studies.

Meta-analysis

Because of the small number of included studies, neither the clinical effects of rTMS and tDCS (n = 2 for each) nor the cognitive effects of rTMS studies (n = 2) and cathodal tDCS (n = 3) could be subjected to meta-analysis. Therefore, we conducted a meta-analysis only on the cognitive effects of anodal tDCS (12 studies) in ADHD. We calculated effect sizes from reported means and standard deviations or t and f values where possible. All data were extracted by S.J.W. We used Plot Digitizer to convert plotted data to numerical values (for an example, see Westwood and Romani³⁵), and we obtained any unreported data by personal communication. All extracted data were cross-checked by K.R.

To reduce large heterogeneity, we clustered cognitive outcome measures into 3 domains and analyzed them in 3 separate meta-analyses. Such clustering of outcome measures into cognitive domains was informed by factor analyses of executive-function measures of ADHD, which typically cluster into factors that comprise measures of attention, inhibition and processing speed.^53–56 Accordingly, for attention measures, we included the numbers or percentages of errors or omission errors (or the inversely reported number or percentage of correct trials) and intrasubject reaction time variability or intrasubject coefficient of variation (intrasubject reaction time variability divided by mean reaction time) to go/congruent/ target trials in go/no-go tasks, flanker tasks, Stroop colour and word tasks (Stroop), working memory tasks and the Wisconsin Card Sorting Task (WCST). For inhibition measures, we included the numbers or percentages of commission errors (or the inversely reported number or percentage of correct trials) to no-go trials in the go/no-go task; the number of percentage errors or reaction time to incongruent trials in flanker or Stroop tasks; the number of commission errors in continuous performance tasks (CPTs) or MOXO tasks; stop signal reaction times in the stop task; perseverative errors in the WCST; and multi-button responses in the MOXO task. For processing speed, we included mean reaction times to go/congruent/target trials in alertness, CPT, go/no-go and MOXO tasks, and completion time in the WCST.

We estimated effect sizes using small-sample-corrected standardized mean differences (i.e., Hedges’ g),⁵⁷ which calculated the difference in performance under sham versus anodal tDCS divided by pooled standard deviation to standardize the effect. We reported effects as positive if a cognitive outcome measure showed improvement with anodal tDCS relative to sham stimulation, and as negative if it showed deterioration. We conducted all meta-analyses using random-effects models to account for heterogeneity (i.e., effect size variation between studies beyond that expected for sampling error alone).⁵⁸ To provide a measure of heterogeneity, we report the I² value; I² values of 75%, 50% and 25% reflect high, moderate and low heterogeneity, respectively.⁵⁹

When estimating Hedges’ g in crossover designs, we accounted for the correlation between pre and post measures (otherwise, there would have been an underestimation of the effect sizes).⁵⁷ Where multiple effects were reported from the same sample, we created composite effect sizes by averaging effect sizes and decreasing variances, assuming correlation across the different effects. Because tDCS studies did not report these correlations, we estimated the crossover correlation from reported t values (25 of 60 reported outcomes) derived from analyses comparing anodal tDCS with sham stimulation, and we estimated the correlation across multiple dependent outcomes from an ongoing neurotherapy intervention study in our laboratory with a sample of 74 adolescents with ADHD tested before and after intervention in the Maudsley Attention and Response Suppression task battery, including go/no-go tasks, CPTs, Simon tasks, time estimation tasks⁵⁴ and the WCST. Specifically, we assumed a correlation of 0.629 between outcome measures for studies with crossover designs, and a correlation of 0.3 between the different effects for composite effects. In addition, we conducted sensitivity analyses assuming crossover correlation of 0.407 and 0.780 (the upper and low 95% CIs of the estimated correlation) and composite correlation of 0.1 and 0.5 to test whether findings depended on our estimates.

Finally, we used jackknife sensitivity analyses (i.e., repeating the same analysis and excluding a different study each time) to establish the replicability of findings. To improve homogeneity, we carried out additional sensitivity analyses that excluded studies with overlapping samples, ^60–63 or studies with methods that deviated from the majority of studies, such as those including community participants with high ADHD symptoms on validated ADHD ratings scales but without a clinical ADHD diagnosis;^43,64 those including adult samples with ADHD;^42,65,66 those targeting the right inferior frontal cortex (IFC);^43,67 those reporting change scores (i.e., post minus pre differences) rather than post scores only;⁶⁶ those using multi-stimulation sessions; ^60–62,65 those with effect sizes based on working memory or WCST tasks;^61,68 or those using parallel rather than crossover designs.^62,66,67 Lastly, given that effect-size estimates might be inflated in studies with a high risk of bias, we conducted meta-regression analyses to compare the effect sizes of studies with high versus low or unclear risk of bias for each risk of bias domain (i.e., selection bias, performance bias, detection bias, attrition bias and other biases). All analyses were conducted by J.R.; meta-analysis calculations were conducted using the function “rma” of the “metafor” package version 2.1 in R version 3.6.^69,70

Results

Literature review

rTMS studies

Two double-blind, crossover studies targeted the right dlPFC. In 13 adults with ADHD, 1 session of 20 Hz rTMS relative to sham significantly improved overall self-rated inattention but not hyperactivity symptoms.⁷¹ In 9 adults with ADHD, 10 daily sessions of 10 Hz rTMS relative to sham showed no effect on self-rated clinical symptoms, nor on electroencephalography or executive-function measures. ⁷² In a single-blind study in 21 adolescents with ADHD, 20 daily sessions of 18 Hz deep rTMS over the bilateral dlPFC (n = 13) compared with sham (n = 9) showed no effect on self-rated clinical or cognitive measures of sustained attention.⁷³ An open-label trial in 10 children with ADHD showed fewer teacher-rated inattention and parentrated hyperactivity/impulsivity symptoms 1 week after 5 daily sessions of 1 Hz rTMS of the left dlPFC compared with baseline.⁵² However, without sham control conditions, placebo effects could not be ruled out (Table 1).

Table 1.

Clinical and cognitive effects of rTMS

Study	Design	N	Age, yr	Region	Stimulation protocol				Outcome measures
					Sessions, n	Frequency, Hz	Intensity, % of MT	Duration	Clinical*	Cognitive
Bloch et al.71	Single-blind, sham-controlled, crossover	13	Adults (age not reported)	Right dlPFC†	1	20	100	1680 pulses (2 s on, 30 s off)	PANAS (inattention [+], total score [+]); VAS (inattention [+])‡	Not tested
Gomez et al.52	Open-label	10	7–12	Left dlPFC	5	1	90	1500 pulses (on/off not reported)	DSM-IV ADHD symptom checklist (parent-rated hyperactivity/impulsivity [+], teacher-rated inattention [+])	Not tested
Paz et al.73	Double-blind, sham-controlled, parallel	Active: 9 Sham: 13	Active: 32 Sham: 30	Bilateral dlPFC§	20	18	120	1980 pulses (2 s on, 20 s off)	CAARS	TOVA
Weaver et al.72	Single-blind, sham-controlled, crossover	9	18	Right dlPFC†	10	10	100	2000 pulses (4 s on, 26 s off)	CGI-I scale; ADHD-RS-IV scale	WASI/WISC-IV; Conners’ CPT; D-KEFS; Buschke Selective Reminding Test; digit symbol coding test; finger oscillation tasks

ADHD = attention-deficit/hyperactivity disorder; ADHD-RS-IV = ADHD Rating Scale, Fourth Edition; CAARS = Conners’ Adult ADHD Rating Scales; CGI-I = Clinical Global Impression–Improvement scale; CPT = continuous performance task; D-KEFS = Delis–Kaplan Executive Function System; dlPFC = dorsolateral prefrontal cortex; DSM-IV = Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition; MT = motor threshold; PANAS = Positive and Negative Affect Schedule; rTMS = repetitive transcranial magnetic stimulation; TOVA = Test of Variables of Attention; VAS = visual analogue scale; WASI = Wechsler Abbreviated Scale of Intelligence (selected subtests from the Wechsler Adult Intelligence Scale); WISC-IV = Wechsler Intelligence Scale for Children, Fourth Edition.

^*Plus sign [+] = statistically significant improvement.

^†5 cm forward to MT point.

^‡Small change from baseline of 0.25 and 1.16 on 5-point Likert scales.

^§6 cm rostral to motor cortex.

tDCS studies

Fourteen studies tested tDCS in ADHD; 9 studies were double-blind, 4 studies were single-blind and 1 study was open-label. Ten studies tested children, and 4 studies tested adults. Only 2 studies combined tDCS with cognitive training (Table 2 and Table 3).

Table 2.

Clinical and cognitive effects of tDCS

Study	Design	N	Mean age, yr	Stimulation protocol					Outcome measures
				Anode/cathode	mA	Sessions, n	Timing*	Duration, min	Clinical†	Cognitive†
Children
Bandeira et al.51‡	Open-label	9	11	Left dlPFC/right SOA	2	5	Online	28	PGI-I	Visual Attention Test (TAVIS-3; omission errors [+]); NEPSY-II-inhibition (switch errors [+]); digit span task; Corsi cube task
Breitling et al.67	Single-blind, sham-controlled, crossover	21	14	Right IFC/left cheek	1	1	Online	20	Not tested	Flanker task (incongruent trials: errors [+]¶,** and ICV [+]¶,††)
				Left cheek/right IFC	1	1	Online	20	Not tested	Flanker task
Munz et al.60	Double-blind, sham-controlled, crossover	14	12	Left dlPFC/right cheek; right dlPFC/left cheek	0.25	1	Offline	25 (5 on, 1 off)	Not tested	Go/no-go task (go RT [+] and intrasubject RTV [+]); motor memory task; alertness task
Nejati et al.68 (experiment 1)	Double-blind, sham-controlled, crossover	15	10	Left dlPFC/right dlPFC	1	1	Offline	15	Not tested	Go/no-go task; n-back (accuracy, RT [+]); Stroop task (incongruent trials: errors [+] and RT [+]); WCST (completion time [+])
Nejati et al.68 (experiment 2)	Double-blind, sham-controlled, crossover	10	9	Left dlPFC/right SOA	1	1	Offline	15	Not tested	Go/no-go task; n-back (accuracy [+],¶ RT [+]**); WCST (total categories completed [+], total errors [+], perseverative errors [+])**
				Right SOA/left dlPFC	1	1	Offline	15	Not tested	Go/no-go task (no-go accuracy [+])**; n-back; WCST (total categories completed [+], total errors [+] and perseverative errors [+]¶)**
Prehn-Kristensen et al.61	Double-blind, sham-controlled, parallel	12	12	Left dlPFC/right cheek; right dlPFC/left cheek	0.25	1	Offline	25 (5 on, 1 off)	Not tested	Declarative memory task (accuracy [+]); alertness task; digit span task
Soff et al.62	Double-blind, sham-controlled, crossover	15	14	Left dlPFC/vertex	1	5	Online	20	FBB-ADHD (inattention [+])‡‡,§§,¶¶	QbTest (inattention [+]††; hyperactivity [+]***)§§,¶¶
Soltaninejad et al.64	Single-blind, sham-controlled, crossover	20	16	Left dlPFC/right SOA	1.5	1	Online	15	Not tested	Go/no-go task (go accuracy [+])¶,**; Stroop task
				Right SOA/left dlPFC	1.5	1	Online	15	Not tested	Go/no-go task (no-go accuracy [+])¶,†††; Stroop task
Soltaninejad et al.43§	Single-blind, sham-controlled, crossover	20	16	Right IFC/left SOA	1	1	Online	15	Not tested	Go/no-go task (go accuracy [+]); Stroop task
Sotnikova et al.63	Double-blind, sham-controlled, crossover	13	14	Left dlPFC/vertex	1	1	Online	20	Not tested	QbTest (RT [+], RTV [+],‡‡‡ omission errors [−], accuracy [−])§§§
Adults
Allenby et al.65‡	Double-blind, sham-controlled, crossover	37	32	Left dlPFC/right SOA	2	3	Online	20	Not tested	Conners’ CPT (commission errors [+]¶¶¶); stop task
Cachoeira et al.74	Double-blind, sham-controlled, parallel	Active: 9 Sham: 8	Active: 31 Sham: 34	Right dlPFC/left dlPFC	2	5	Offline	20	ADHD checklist (inattention [+], total [+])****; SDS (after tDCS only [+])	Not tested
Cosmo et al.66	Double-blind, sham-controlled, parallel	Active: 30 Sham: 30	Active: 32 Sham: 33	Left dlPFC/right dlPFC	1	1	Offline	20	Not tested	Go/no-go task
Jacoby et al.42	Single-blind, sham-controlled, crossover	20	23	Left and right dlPFC/cerebellum	1.8	1	Offline	20	Not tested	CPT (multi-button presses [+])

ADHD = attention deficit/hyperactivity disorder; CPT = continuous performance task; dlPFC = dorsolateral prefrontal cortex; FBB-ADHD = parents’ version of a German adaptive diagnostic checklist for ADHD; ICV = intrasubject coefficient of variation; IFC = inferior frontal cortex; NEPSY-II = Neuropsychological Development Assessment, Second Edition; PGI-I = Patient Global Impression of Improvement; QbTest = Quantitative Behaviour Test; RT = reaction time; RTV = reaction time variability or standard deviation of reaction times; SDS = Sheehan Disability Scale; SOA = supraorbital area; Stroop = Stroop colour and word task; tDCS = transcranial direct current stimulation; WCST = Wisconsin Card Sorting Task.

^*Refers to whether cognitive performance was during stimulation (online) or after stimulation (offline).

^†Plus sign [+] = statistically significant improvement; minus sign [−] = statistically significant impairment.

^‡Combined stimulation with cognitive training.

^§Originally published in Persian; translated by the lead author (ZS).

^¶Would likely not survive multiple comparison correction.

^**Comparisons between stimulation conditions based on post-hoc least significant difference tests, which do not correct for multiple comparisons.

^††Based on underpowered analysis focusing on the first session, with 7 participants per condition.

^‡‡Improvement seen only 7 days after the fifth anodal tDCS session.

^§§Did not survive correction for multiple comparisons.

^¶¶Based on underpowered analysis focusing on the first 5 sessions, with 7/8 participants per condition.

^***Improvement seen immediately after the fifth anodal tDCS session and 7 days later.

^†††Significant in comparison to cathodal tDCS only.

^‡‡‡Based on a crossover interaction; tDCS reduced RT and RTV in 1 of 4 conditions (2-back tasks), but this did not survive correction for multiple comparisons.

^§§§Included carryover effect raised by Soff et al.⁶²

^¶¶¶Significant only immediately after anodal tDCS; not significant 3 days later.

^****Inattention improved immediately after anodal tDCS and after 2 weeks; total score improved only after 2 weeks.

Table 3.

Cognitive task measures (mean ± SD) extracted from studies

Study	Cognitive task	Cognitive outcome measure	tDCS, mean ± SD	Sham, mean ± SD	Effect*
Measures of attention
Allenby et al.65	CPT	No. omission errors	1.9 ± 4.3	2.1 ± 2.4	Single
Breitling et al.67	Flanker task (incongruent trials)	% Omission errors	2.2 ± 4.6	5.8 ± 7.6	Composite
	Flanker task (incongruent trials)	Intraindividual coefficient of variation (ms)†	0.2 ± 0.1	0.3 ± 0.1
Cosmo et al.66	Go/no-go task (fruits)	No. omission errors (post/pre)‡	−2.9 ± 24.5	−3.7 ± 21.2	Single
Jacoby et al.42	MOXO task	RT (ms) with 1 distractor	552.2 ± 53.9	558.9 ± 52.1	Composite
	MOXO task	RT (ms) with 2 distractors	556.5 ± 54.4	568.9 ± 53.5
	MOXO task	No. omission errors	4.2 ± 4.1	4.3 ± 4.0
Munz et al.60	Alertness task	Intrasubject RTV (ms)	69.5 ± 25.1	76.1 ± 31.2	Composite
	Go/no-go task (go trials)	Intrasubject RTV (ms)	225.2 ± 246.9	379.4 ± 425.3
	Go/no-go task (go trials)	No. omission errors	3.0 ± 3.0	4.3 ± 4.4
Nejati et al.68 (experiment 1)	Go/no-go task (go trials)	% Correct	93.3 ± 11.4	90.9 ± 19.6	Composite
	N-back task (1-back)	No. correct	15.3 ± 8.3	14.4 ± 7.4
	WCST	No. of categories completed	2.5 ± 0.8	2.4 ± 1.1
	WCST	No. total errors	29.7 ± 8.3	30.7 ± 9.3
Nejati et al.68 (experiment 2)	Go/no-go task (go trials)	% Correct	100 ± 0.0	98.5 ± 3.2	Composite
	N-back task (1-back)	No. correct	21.0 ± 2.3	17.9 ± 2.9
	WCST	No. of categories completed	5.1 ± 0.7	3.9 ± 0.7
	WCST	No. total errors	11.0 ± 2.9	21.6 ± 5.4
Prehn-Kristensen et al.61	Digit span task	No. correct	9.6 ± 2.4	10.8 ± 2.1	Single
Soff et al.62	QbTest (inattention)	z-scores (omission errors, RT and intrasubject RTV)	0.1 ± 1.2	−0.1 ± 0.4	Single
Soltaninejad et al.64	Go/no-go task (go trials)	% Correct	98.8 ± 3.6	98.9 ± 1.9	Single
Soltaninejad et al.43	Go/no-go task (go trials)§	% Correct	—	—	Single
Sotnikova et al.63	QbTest (overall)	Intrasubject RTV (ms)	214.3 ± 97.2	235.2 ± 122.7	Composite
	QbTest (overall)	No. omission errors	38.6 ± 22.8	22.5 ± 15.3
	QbTest (overall)	% Correct¶	31.7 ± 5.1	43.5 ± 6.9
Measures of inhibition
Allenby et al.65	CPT	No. commission errors	17.1 ± 9.1	19.8 ± 10.9	Composite
	Stop task	Stop signal RT	288.4 ± 76.0	291.5 ± 68.1
Breitling et al.67	Flanker task (incongruent trials)	RT (ms)	581.0 ± 43.0	585.0 ± 38.0	Composite
	Flanker task (incongruent trials)	% Errors	9.8 ± 7.2	20.6 ± 9.2
Cosmo et al.66	Go/no-go task (fruits)	No. commission errors (post/pre)‡	−5.5 ± 10.0	−6.9 ± 10.4	Single
Jacoby et al.42	MOXO task	Multi-button responses	4.7 ± 4.9	7.0 ± 6.3	Composite
	MOXO task	No. commission errors	11.3 ± 11.2	11.7 ± 12.1
Munz et al.60	Go/no-go task (no-go trials)	No. commission errors	15.7 ± 10.3	12.6 ± 8.2	Single
Nejati et al.68 (experiment 1)	Go/no-go task (no-go trials)	% correct	19.9 ± 7.6	19.0 ± 7.8	Composite
	Stroop task (incongruent trials)	RT (ms)	2870 ± 2210	1390 ± 440
	Stroop task (incongruent trials)	% Errors	24.9 ± 12.0	34.9 ± 15.5
	WCST	No. perseverative errors	17.6 ± 3.6	18.0 ± 9.0
Nejati et al.68 (experiment 2)	Go/no-go task (no-go trials)	% Correct	22.7 ± 1.3	20.7 ± 4.4	Composite
	WCST	No. perseverative errors	7.8 ± 2.4	14.8 ± 3.7
Soff et al.62	QbTest (impulsivity)	z-scores (commission errors, multi-button press per stimulus, anticipatory button press)	0.2 ± 1.2	−0.2 ± 0.7	Single
Soltaninejad et al.64	Go/no-go task (no-go trials)	% Correct	96.2 ± 8.2	95.8 ± 6.9	Composite
	Stroop task (incongruent trials)	% Correct	98.3 ± 2.9	96.4 ± 3.6
	Stroop task (incongruent trials)	RT (ms)	1080 ± 180	1130 ± 220
Soltaninejad et al.43	Go/no-go task (no-go trials)§	% Correct	—	—	Composite
	Stroop task (incongruent trials)§	% Correct	—	—
	Stroop task (incongruent trials)§	RT (ms)	—	—
Sotnikova et al.63	QbTest (overall)	No. commission errors	6.0 ± 5.1	6.0 ± 4.2	Single
Measures of processing speed
Allenby et al.65	CPT	RT (ms) to target	420.9 ± 63.3	419.7 ± 73.0	Single
Jacoby et al.42	MOXO task	RT (ms) to target	541.5 ± 50.8	547.0 ± 53.5	Composite
Munz et al.60	Alertness task	RT (ms)	309.6 ± 51.8	302.4 ± 44.3	Composite
	Go/no-go task (go trials)	RT (ms)	453.2 ± 131.3	566.9 ± 234.1
Nejati et al.68 (experiment 1)	Go/no-go task (go trials)	RT (ms)	1080 ± 210	1030 ± 170	Composite
	N-back task (1-back)	RT (ms)	120.2 ± 22.5	175.7 ± 55.4
	WCST	Completion time (ms)	237 300 ± 79 800	291 100 ± 106 700
Nejati et al.68 (experiment 2)	Go/no-go task (go trials)	RT (ms)	1330 ± 900	1230 ± 120	Composite
	N-back task (1-back)	RT (ms)	103.4 ± 24.2	162.9 ± 94.4
	WCST	Completion time (ms)	123 200 ± 16 900	170 300 ± 85 900
Soltaninejad et al.64	Go/no-go task (go trials)	RT (ms)	830 ± 290	910 ± 350	Single
Soltaninejad et al.43	Go/no-go task (go trials)§	RT (ms)	—	—	Single
Sotnikova et al.63	QbTest (overall)	RT (ms)	555.3 ± 116.2	564.2 ± 130.8	Single

CPT = continuous performance task; QbTest = Quantitative Behaviour Test; RT = reaction time; RTV = reaction time variability; SD = standard deviation of the mean; Stroop = Stroop colour word task; WCST = Wisconsin Card Sorting Task.

^*For multiple effects from the sample, we created composite effect-size estimates; single effects otherwise.

^†Intrasubject reaction time variability divided by mean reaction time.

^‡Only change scores (post – pre/baseline) were reported.

^§The author could not provide raw means and standard deviations for each stimulation condition, so to calculate Hedges’ g we converted the reported t-statistics.⁵⁷

^¶The % correct was reported as: hits (total no. of target trials – no. of omission errors errors) + correct rejections (total no. of no-go trials – no. of commissions errors)/total number of stimuli⁶³

tDCS studies in children with ADHD

Two single-blind, sham-controlled crossover studies in 20 high school students who scored above the clinical cut-off for ADHD on validated ADHD questionnaires applied a single session of anodal tDCS over the left dlPFC⁶⁴ or the right IFC.⁴³ Anodal relative to cathodal tDCS of the left dlPFC improved go accuracy, and cathodal tDCS improved no-go accuracy compared with anodal tDCS and sham treatment, but none improved performance on the Stroop task.⁶⁴ Right IFC anodal tDCS relative to sham treatment improved go accuracy, but there were no equivalent improvements on any other go/no-go or Stroop task measures.⁴³

Two double-blind, sham-controlled, crossover studies applied single-session stimulation over the dlPFC. In 15 adolescents with ADHD, tDCS over the bilateral dlPFC (anode left/cathode right) improved WCST completion time and n-back and Stroop reaction times and errors to incongruent trials, but had no effects on n-back accuracy or go/no-go measures. Furthermore, errors and reaction times to incongruent trials in the Stroop task were used as main outcome measures, rather than the Stroop interference or error effect (reaction time/errors on Stroop – reaction time/errors on congruent trials) — the established key outcome measure of the Stroop task. In 10 adolescents with ADHD, anodal tDCS of the left dlPFC improved n-back working memory accuracy and reaction times compared to both sham treatment and cathodal tDCS. Both anodal and cathodal tDCS over the left dlPFC improved WCST performance; anodal tDCS had stronger effects. Cathodal tDCS of the left dlPFC also improved no-go accuracy, presumably via interhemispheric inhibition increasing right prefrontal regions,⁶⁸ which have been associated with motor response inhibition in children and adults.^54,75 This explanation is partly supported by a single-blind crossover study in 21 adolescents with ADHD, which found that 1 session of anodal (but not cathodal) tDCS over the right IFC versus sham treatment reduced errors (trend level) and reaction time variability in a flanker task.⁶⁷ However, as in the previous paper,⁶⁸ findings were based on flanker incongruent trials rather than on flanker interference reaction time or error effect, and the analysis included only the first session to remove a practice effect, reducing the sample size to 7 participants per condition.

One double-blind, crossover study applied 5 daily sessions of anodal versus sham tDCS over the left dlPFC in 15 adolescents with ADHD. The results showed improvements relative to sham treatment in parent-rated clinical measures of inattention and on cognitive measures of attention (as assessed using the QbTest, a combined working memory and go/no-go task) 7 days after anodal tDCS but not immediately after, and improvements in QbTest measures of hyperactivity both immediately after anodal tDCS and 7 days later. Clinical and QbTest impulsiveness measures showed no effects. However, findings were based on the first 5 sessions to remove a carryover effect, reducing the sample to 7 or 8 participants per condition. ⁶² In the same study, 13 of the 15 adolescents with ADHD showed reduced reaction time variability but increased errors on the QbTest after a single session of anodal tDCS. However, this analysis included a carryover effect.⁶³

A double-blind, sham-controlled crossover study found that relative to sham treatment, overnight slow-wave oscillatory anodal tDCS over the left and right dlPFC improved declarative memory in 12 children with ADHD⁶¹ and go reaction time and its intrasubject variability in 14 children with ADHD,⁶⁰ but had no effects on no-go accuracy, or on measures of alertness, digit span or motor memory.

The only open-label trial in 9 children with ADHD found that 5 daily sessions of anodal tDCS to the left dlPFC combined with a picture association cognitive training task reduced errors on attention (omission) and switch tasks but did not improve working memory. Parents reported improvements in some of their children’s behaviour except for one, who reported their child was “much worse.” However, without a sham control condition, findings could have been confounded by placebo, test–retest or cognitive training effects.⁵¹

tDCS studies in adults with ADHD

A double-blind, parallel study found no effects of a single session of anodal tDCS over the left dlPFC relative to sham treatment on go/no-go performance in 60 adults with ADHD.⁶⁶ A single-blind, crossover study in 20 undergraduates with ADHD found that 1 session of bifrontal anodal tDCS over the left and right dlPFC relative to sham treatment improved hyperactivity measures in a sustained attention task (i.e., multiple/random responses), but not omission errors or reaction times.⁴² A double-blind, crossover study in 37 adults with ADHD found that 3 sessions over alternate days of visual working memory training combined with anodal tDCS relative to sham tDCS of the left dlPFC reduced commission errors in a sustained attention task immediately after anodal tDCS, but not at the 3-day follow-up; there were no effects on omission errors, reaction times or stop task performance immediately after anodal tDCS or 3 days later.⁶⁵ Finally, a double-blind, parallel study in 17 adults with ADHD reported that 5 daily sessions of anodal right tDCS (n = 9) relative to sham treatment (n = 8) improved inattention but not hyperactivity/impulsive symptoms immediately after stimulation and 2 weeks later, at which point the total ADHD score was also improved.

Safety in tDCS studies

Brain stimulation was well tolerated overall; the most commonly reported side effects were mild tingling and itching. One study reported dropouts because of tingling sensations and headache⁴³ (Table 4). There was 1 case of an adverse effect, where 1 patient reported hypobulia after a single session of anodal tDCS over the right dlPFC, which persisted in a milder form the following day.⁷⁴

Table 4.

Side effects and adverse effects reported in studies using rTMS and tDCS

Study	Side and adverse effects
rTMS studies
Bloch et al.71	Not tested, but authors observed differences in the somatosensory experience of real rTMS and sham, which limited true blinding
Gomez et al.52	Majority reported transient mild headache and scalp discomfort; a minority reported neck pain. No seizure-like cortical activity was found. One case of slight and brief dizziness
Paz et al.73	Not tested
Weaver et al.72	Minority reported transient mild headaches and scalp discomfort
tDCS studies
Allenby et al.65	Burning, itching, pain and tingling were rated significantly higher during anodal tDCS compared with sham
Bandeira et al.51	Majority reported mild to moderate headache, tingling, itching, burning sensation and redness of the skin. Minority reported mild neck pain, sleepiness and static shock. One parent reported child was “much worse” after stimulation
Breitling et al.67	Skin sensations were rated higher during cathodal tDCS relative to sham
Cachoeira et al.74	Comparable reports of headache, tingling, itching, burning sensation and tiredness were found in both sham and anodal tDCS. One patient withdrew after the first sessions because of an “acute mood change, feeling sad, hypobulia, tension … 5 hours after stimulation and persisted in a milder form into the next day”
Cosmo et al.66	None reported
Jacoby et al.42	Not tested, but authors reported that stimulation was well tolerated and no side effects were reported
Munz et al.60	Not tested, but no side effects were reported
Nejati et al.68	Reports of mild itching or tingling under electrodes
Prehn-Kristensen et al.61	Not tested. No side effects reported
Soff et al.62	About half of participants reported tingling and itching sensations under electrode during sham and anodal tDCS
Soltaninejad et al.64*	Not reported
Soltaninejad et al.43*	Several participants dropped out because of tingling sensation and headache
Sotnikova et al.63	About half of participants reported tingling and itching sensations under electrode during sham and anodal tDCS

rTMS = repetitive transcranial magnetic stimulation; tDCS = transcranial direct current stimulation.

^*Personal communication, 2019.

Summary of findings of tDCS studies

Findings were mixed. With respect to the clinical effects of tDCS, only 2 sham-controlled studies tested and found improvement in behavioural symptoms of inattention but not in impulsiveness/hyperactivity symptoms, and an openlabel trial found improvements in parent-rated impression of improvement in global functioning. Of 14 studies that tested cognitive effects, 13 observed positive effects on some cognitive functions but not others, and 1 reported worse performance in a sustained attention task.⁶³ Moreover, 7 sham-controlled studies did not correct for multiple comparisons, ^{43,62–64,67,68} and the majority of their findings would not have survived correction.

Risk of bias

A minority of studies indicated unclear bias for selection (n = 6), detection (n = 1) and attrition (n = 2), but on balance bias was low in these domains. Performance bias (i.e., blinding participants and personnel) was unclear in 11 studies and high in 2 studies; selective reporting bias and other biases (e.g., nonstandard outcome measures, no correction for multiple testing, lenient significance threshold) were high in 4 studies (Fig. 2; Appendix 1, Table S2).

Fig. 2.

Risk of bias ratings for (A) repetitive transcranial magnetic stimulation (rTMS), (B) transcranial direct current stimulation (tDCS) and (C) individual studies. Note: risk of bias ratings were the same for both studies reported in Nejati et al.⁶⁸

Meta-analysis of anodal tDCS studies

The majority of the studies included in the meta-analysis (10 of 12) used anodal stimulation of mostly left dlPFC regions (unilateral left, n = 5; bifrontal anode left and right, n = 3; bilateral anode left/cathode right, n = 2); only 2 studies involved unilateral stimulation of the right IFC (Table 2). The meta-analyses showed no significant effect in measures of attention (Hedges’ g = 0.18, 95% CI −0.19 to 0.45, p = 0.20) with high heterogeneity (I² = 75%, p < 0.001) and trend-level improvements in measures of inhibition (Hedge’s g = 0.21, 95% CI −0.01 to 0.43, p = 0.06) with moderate heterogeneity (I² = 60%, p = 0.01) and of processing speed (Hedges’ g = 0.14, 95% CI −0.01 to 0.29, p = 0.07) with low heterogeneity (I² = 3%, p = 0.50; Fig. 3). Sensitivity analyses confirmed that these findings did not systematically rely on our estimated correlations for effects from crossover studies or for composite effect-size estimates (Appendix 1, Table S3). Further, we replicated these findings in the jackknife sensitivity analyses (i.e., repeating the main analyses but excluding a different study with each repetition), with only minor exceptions: excluding a minority of individual studies led to a significant but still small effect in measures of inhibition (excluding Cosmo and colleagues,⁶⁶ Munz and colleagues⁶⁰ and Soltaninejad and colleagues⁴³) and in processing speed (excluding Allenby and colleagues⁶⁵ and Soltaninejad and colleagues;⁴³ Table 5).

Fig. 3.

Meta-analysis of measures of (A) attention, (B) inhibition and (C) processing speed. CI = confidence interval; tDCS = transcranial direct current stimulation.

Table 5.

Results of jackknife sensitivity analysis

Meta-analysis	Studies excluded	Effect size			Heterogeneity
		Hedges’ g*	95% CI	p value	I2	p value
Attention	Allenby et al.65	0.19	−0.11 to 0.50	0.22	77	< 0.001
	Breitling et al.67	0.14	−0.13 to 0.42	0.31	76	< 0.001
	Cosmo et al.66	0.20	−0.10 to 0.50	0.19	78	< 0.001
	Jacoby et al.42	0.19	−0.12 to 0.49	0.24	77	< 0.001
	Munz et al.60	0.16	−0.14 to 0.46	0.30	77	< 0.001
	Nejati et al.68 (experiment 1)	0.19	−0.12 to 0.50	0.22	77	< 0.001
	Nejati et al.68 (experiment 2)	0.08	−0.12 to 0.27	0.43	49	0.040
	Prehn-Kristensen et al.61	0.23	−0.03 to 0.50	0.09	72	0.001
	Soff et al.62	0.18	−0.11 to 0.47	0.22	78	< 0.001
	Soltaninejad et al.64	0.20	−0.10 to 0.50	0.19	77	< 0.001
	Soltaninejad et al.43	0.14	−0.15 to 0.42	0.35	75	0.001
	Sotnikova et al.63	0.24	−0.03 to 0.50	0.08	70	0.002
Inhibition	Allenby et al.65	0.22	−0.04 to 0.48	0.09	65	0.010
	Breitling et al.67	0.19	−0.03 to 0.41	0.10	62	0.010
	Cosmo et al.66	0.24	0.01 to 0.48	0.040	63	0.010
	Jacoby et al.42	0.20	−0.05 to 0.44	0.12	65	0.010
	Munz et al.60	0.25	0.05 to 0.46	0.020	53	0.020
	Nejati et al.68 (experiment 1)	0.19	−0.06 to 0.43	0.13	63	0.010
	Nejati et al.68 (experiment 2)	0.15	−0.03 to 0.32	0.10	37	0.12
	Soff et al.62	0.21	−0.02 to 0.44	0.07	65	0.010
	Soltaninejad et al.64	0.20	−0.05 to 0.45	0.12	65	0.010
	Soltaninejad et al.43	0.26	0.04 to 0.48	0.020	52	0.030
	Sotnikova et al.63	0.23	−0.01 to 0.47	0.06	66	0.010
Processing speed	Allenby et al.65	0.19	0.02 to 0.35	0.030	0	0.53
	Jacoby et al.42	0.15	−0.02 to 0.32	0.09	13	0.38
	Munz et al.60	0.16	0.00 to 0.33	0.06	10	0.44
	Nejati et al.68 (experiment 1)	0.09	−0.08 to 0.25	0.29	0	0.79
	Nejati et al.68 (experiment 2)	0.12	−0.04 to 0.28	0.15	2	0.50
	Soltaninejad et al.64	0.13	−0.04 to 0.29	0.14	9	0.42
	Soltaninejad et al.43	0.17	0.01 to 0.33	0.040	5	0.50
	Sotnikova et al.63	0.15	−0.02 to 0.31	0.08	11	0.40

CI = confidence interval.

^*Effect size estimates assumed a correlation of 0.629 for crossover studies and a correlation of 0.3 for multiple dependent effects.

To improve homogeneity, we carried out additional sensitivity analyses to exclude studies with overlapping samples, methods that departed from the majority of studies or studies that were of lower methodological quality (Table 6). For inhibition measures, these analyses revealed significant but small effects when the analysis was limited to studies using single-session tDCS (Hedges’ g = 0.28, 95% CI 0.01 to 0.56, p = 0.04); reported only post-stimulation scores (Hedges’ g = 0.24, 95% CI 0.01 to 0.48, p = 0.04); or stimulated mainly the left dlPFC (Hedges’ g = 0.24, 95% CI 0.01 to 0.47, p = 0.04), all of which were associated with significant moderate heterogeneity (I² = 66, p = 0.01; I² = 63, p = 0.01; I² = 56, p = 0.02, respectively). For processing speed measures, we found significant but small effects when the analysis was limited to studies that used single- session tDCS (Hedges’ g = 0.22, 95% CI 0.04 to 0.41, p = 0.02), used child samples (Hedges’ g = 0.20, 95% CI 0.01 to 0.39, p = 0.04) or stimulated mainly the left dlPFC (Hedges’ g = 0.17, 95% CI 0.01 to 0.33, p = 0.04), all of which were associated with low heterogeneity (I² = 0, p = 0.51; I² = 9, p = 0.41; I² = 5, p = 0.50, respectively).

Table 6.

Results for sensitivity analysis of the meta-analyses of attention, inhibition and processing speed

Included studies	Excluded studies	Studies	Effect sizes			Heterogeneity
			Hedges’ g	95% CI	p value	I2	p value
Attention
All	None	12	0.18	−0.19 to 0.45	0.20	75	< 0.001
Studies with crossover design	Allenby et al.65; Breitling et al.67*; Soff et al.62	9	0.17	−0.17 to 0.50	0.32	82	< 0.001
Patients with clinical ADHD diagnosis	Soltaninejad et al.43; Soltaninejad et al.64†	10	0.16	−0.16 to 0.48	0.33	78	< 0.001
No WCST and working memory tasks	Nejati et al.68‡; Prehn-Kristensen et al.61	11	0.14	−0.04 to 0.32	0.13	39	0.12
No overlapping samples§	Munz et al.60; Soff et al.62	10	0.16	−0.16 to 0.49	0.32	80	< 0.001
	Munz et al.60; Sotnikova et al.63	10	0.22	−0.08 to 0.53	0.15	74	0.002
	Prehn-Kristensen et al.61; Soff et al.62	10	0.24	−0.04 to 0.52	0.09	76	< 0.001
	Prehn-Kristensen et al.61; Sotnikova et al.63	10	0.29	0.05 to 0.54	0.020	62	0.010
Only single-session tDCS	Allenby et al.65¶; Munz et al.60; Prehn-Kristensen et al.61; Soff et al.62	8	0.26	−0.11 to 0.64	0.17	81	< 0.001
Report of post scores only	Cosmo et al.66**	11	0.20	−0.10 to 0.50	0.19	78	< 0.001
Target site of dlPFC only	Breitling et al.67††; Soltaninejad et al.43	10	0.10	−0.19 to 0.39	0.51	76	0.001
Only studies in children	Allenby et al.65; Cosmo et al.66‡‡; Jacoby et al.42	9	0.24	−0.16 to 0.63	0.24	80	< 0.001
Inhibition
All	None	11	0.21	−0.01 to 0.43	0.06	60	0.01
Studies with crossover design	Breitling et al.67*; Cosmo et al.66; Soff et al.62	8	0.22	−0.04 to 0.48	0.09	71	0.004
Patients with clinical ADHD diagnosis	Soltaninejad et al.43; Soltaninejad et al.64†	9	0.26	−0.02 to 0.53	0.07	62	0.02
No WCST and working memory tasks	Nejati et al.68‡	11	0.17	0.00 to 0.35	0.06	41	0.09
Only single-session tDCS	Allenby et al.65¶; Munz et al.60; Soff et al.62	8	0.28	0.01 to 0.56	0.040	66	0.01
Report of post scores only	Cosmo et al.66**	10	0.24	0.01 to 0.48	0.040	63	0.01
Target site of dlPFC only	Breitling et al.67††; Soltaninejad et al.43	9	0.24	0.01 to 0.47	0.040	56	0.02
Studies in children only	Allenby et al.65; Cosmo et al.66‡‡; Jacoby et al.42	8	0.26	−0.08 to 0.60	0.13	72	0.004
Processing speed
All	None	8	0.14	−0.01 to 0.29	0.07	3	0.50
Patients with clinical ADHD diagnosis	Soltaninejad et al.43; Soltaninejad et al.64†	6	0.16	−0.02 to 0.34	0.09	13	0.40
No WCST and working memory tasks	Nejati et al.68‡	8	0.00	−0.16 to 0.16	1.00	0	0.78
Only single-session tDCS	Cosmo et al.66¶; Munz et al.60	6	0.22	0.04 to 0.41	0.020	0	0.51
Target site of dlPFC	Soltaninejad et al.43††	7	0.17	0.01 to 0.33	0.040	5	0.50
Studies in children	Allenby et al.65‡‡; Jacoby et al.42	6	0.20	0.01 to 0.39	0.040	9	0.41

ADHD = attention deficit/hyperactivity disorder; CI = confidence interval; dlPFC = dorsolateral prefrontal cortex; IFC = inferior frontal cortex; tDCS = transcranial direct current stimulation; WCST = Wisconsin Card Sorting Task.

^*These sudies used a parallel design; all others used crossover designs.

^†Participants did not have a clinical diagnosis of ADHD.

^‡Only WCST and working memory data excluded.

^§To remove overlapping samples, we excluded 1 study at a time.

^¶These studies applied tDCS over multiple sessions; all others used one session.

^**The only studies to use change scores (post/pre stimulation scores), all other used post-tDCS scores.

^††The only studies stimulated the right IFC; all other targeted the dlPFC.

^‡‡Included adults; all others included children and adolescents.

Finally, meta-regression analyses compared the effect sizes from studies with a high risk of reporting bias versus the effect sizes from studies with low or unclear risk of such bias. At a descriptive level, the effect sizes from studies with high risk of reporting bias or “other” biases (e.g., no correction to multiple testing, lenient α level) were larger than the effect sizes from studies with low or unclear risk. These differences were not statistically significant in any of the outcomes for reporting bias (lowest p = 0.14), but “other” biases yielded a significant effect in measures of inhibition (p = 0.03) and processing speed (p = 0.03). Because none of the studies showed high risk of selection, detection or attrition bias, and only 1 study showed high risk of performance bias, we did not carry out meta-regression analyses for these biases.

Discussion

There has been a recent proliferation of noninvasive brain stimulation studies in ADHD, with large heterogeneity in methodology and outcome measures. We therefore conducted a systematic review of rTMS and tDCS studies. Furthermore, we conducted a rigorous meta-analysis of tDCS studies in ADHD that clustered cognitive effects into clearly separated cognitive domains to increase homogeneity, controlled for multiple dependent effects and potential bias of studies, and tested the replicability of meta-analysis results with jackknife and other sensitivity analyses. The meta-analysis of tDCS studies of mostly the dlPFC showed only trend-level effects on improvement of inhibition and processing speed, but not on attention, suggesting limited effects of tDCS on cognition in children and adults with ADHD.

The findings of the systematic review revealed that for rTMS, of the 4 included studies, only 2 measured and found clinical effects, while the other 2 measured and found no effects on cognitive functions. For tDCS, 12 of the 14 studies included in the systematic review stimulated the dlPFC with anodal tDCS, mostly over the left hemisphere (unilateral left, n = 6; bilateral anode left/cathode right, n = 2; bifrontal left and right, n = 3), and 1 study applied bilateral cathodal tDCS over the left dlPFC (anode right/ cathode left), and 2 studies used unilateral anodal tDCS over the right IFC. Only 2 sham-controlled tDCS studies measured clinical outcomes and found that anodal tDCS over the left or right dlPFC improved ADHD inattention, but not impulsiveness/hyperactivity symptoms, immediately after tDCS and 1 week^62,74 or 2 weeks later.⁷⁴ This could suggest that dlPFC stimulation can improve clinical inattention in ADHD, but this finding will need to be confirmed by future, larger studies.

Importantly, the rigorous meta-analysis of 12 tDCS studies applying 1 to 5 sessions of anodal tDCS of mostly left or bilateral/bifrontal dlPFC (with the exception of Soltaninejad and colleagues⁴³ and Breitling and colleagues⁶⁷) showed only trend-level improvements with small effect sizes in inhibition and processing speed, but no effects on attention.

This systematic review of tDCS and rTMS studies showed that there is very limited evidence that rTMS and tDCS can have an effect on clinical symptoms based on a small sample of 3 sham-controlled studies. The meta-analyses showed — albeit with small effect sizes — that 1 to 5 sessions of anodal tDCS over mostly the left (but also bilateral and bifrontal) dlPFC improved performance on cognitive measures of inhibition and processing speed, but there was no evidence for improvement in attention measures.

Given that 11 of 12 sham-controlled studies in the systematic review tested and found tDCS-induced improvement in 1 or more executive-function measures,^{42,43,60–65,67,68} the findings of the meta-analysis may seem disappointing. However, the lack of positive overall findings, despite positive findings in individual studies, was likely due to underpowered small study effects that were uncorrected for multiple testing^{43,62– 64,67,68} and did not survive the scrutiny of meta-analysis. The decision to cluster cognitive outcome measures into domains for the meta-analysis introduced additional heterogeneity, particularly in the attention and inhibition domains; the processing speed domain was more homogeneous.

The relatively small trend-level effect on inhibition may have been related to the fact that predominantly the right hemispheric IFC mediates inhibitory functions in children and adults,^74,76–84 and that meta-analyses of fMRI studies in ADHD show underactivation of the right IFC during inhibition tasks.^5,6 Stimulation of the right IFC may be more effective for improving inhibitory functions in ADHD. This hypothesis is supported by studies in healthy adults showing that right IFC stimulation improves inhibitory performance. ^85–88 The meta-analysis of Salehinejad and colleagues⁴¹ reported a similarly small, albeit significant, benefit (Hedges’ g = 0.12) with anodal tDCS of mostly the dlPFC on inhibitory measures, which survived when the analysis focused on accuracy measures or only on studies stimulating the dlPFC. However, Salehinejad and colleagues⁴¹ included multiple dependent effects in their meta-analyses, which could have reduced variation between effect sizes and therefore overestimated significance.^44,45 Further, their meta-analysis included noninhibitory measures of attention (e.g., CPT omission errors), processing speed (e.g., reaction times to go trials) and reaction time variability,⁴¹ and therefore was not specific to inhibitory control, which could explain the differences in findings compared to our meta-analysis.

Although right IFC has been more clearly implicated in motor response inhibition, the small, trend-level improvement in inhibition in our meta-analyses, in line with the small significant effect of Salehinejad and colleagues,⁴¹ may have been due to the fact that a majority of inhibitory measures were derived from interference inhibition tasks, which have been shown to be co-mediated by the left dlPFC and IFC⁸⁹ — in particular the Stroop task.^90–93 Further, tDCS studies in healthy adults have reported improved performance on interference inhibition tasks following anodal tDCS of the left dlPFC.^94–97 Given that only 2 studies stimulated the right IFC, future studies will have to test whether inhibition functions can be enhanced with anodal tDCS over the right IFC, which would be in line with neuroimaging evidence of a role of the right IFC in inhibitory control^{76,78,79,81,82,98–101} and evidence of its underactivation in ADHD,^5,6,9 as well as with tDCS studies in healthy adults showing that right IFC stimulation improves motor inhibitory performance.^85–88

The trend-level positive effect of tDCS of mostly the left dlPFC on processing speed is in line with fMRI evidence indicating that the left dlPFC is a key mediating region of processing speed.^102–104 Interestingly, when the processing speed analysis excluded the only study stimulating the right IFC⁴³ — and therefore included only studies stimulating mainly the left dlPFC (unilateral left, n = 5; bilateral, n = 1; bifrontal anodal left and right, n = 2)—the improvement in processing speed became significant, in line with neuroimaging evidence that the left dlPFC mediates processing speed and suggesting that the left dlPFC is an optimal site for enhancing processing speed in ADHD. These findings are partly in line with those of the meta-analysis by Salehinejad and colleagues, ⁴¹ which showed that predominantly anodal tDCS over the left dlPFC improved reaction times and its intrasubject variability in n-back tasks.

The lack of systematic positive effects on attention measures may have been because predominantly the right dlPFC and IFC mediate attention, as evidenced by individual studies and meta-analyses of fMRI studies in children and adults,^{77–79,84,105–107} and in meta-analyses of fMRI studies in ADHD, which show functional underactivation of the right dlPFC/IFC during attention tasks.⁶ It is therefore possible that stimulation of the right dlPFC and/or right IFC may be more effective for improving attention functions in ADHD.

This meta-analysis of tDCS studies shows that anodal tDCS of mostly left dlPFC has only very limited, trend-level effects on improving inhibition and processing speed, with no evidence for attention improvement. However, we cannot rule out the possibility that stimulation of other prefrontal regions (such as the right hemispheric IFC or dlPFC or parietal regions), multiple session tDCS or tDCS in combination with cognitive training could improve clinical or cognitive functions in ADHD.

With respect to safety, stimulation was well tolerated overall, but 1 tDCS study reported higher errors on a sustained attention task⁶³ and another study reported a hypobulia episode in 1 patient,⁷⁴ raising neuroethical concerns of potential costs to nontargeted functions.^38,108 It has been shown that stimulation of a particular region could impair functions mediated by other regions such as the homologue contralateral region via interhemispheric inhibition or other regions that are top-down controlled by the stimulated region.^109,110

Future studies need to address the lack of knowledge about optimal stimulation protocols for children with ADHD. Current knowledge about stimulation effects on the brain and standard protocols are largely derived from adult samples and are therefore not appropriate for children.^108,111 In healthy adults, multi-session stimulation combined with cognitive training may lead to longer-term effects,^30,112 but we do not know whether this protocol can lead to maladaptive plasticity in the developing brain, especially during “sensitivity periods,” where use-dependent plasticity changes are strongest¹¹³ and the possibility of longer-term side effects when using noninvasive brain stimulation in pediatric samples needs more empirical investigation. Future studies should heed recommendations to comprehensively assess effects in children to capture possible unintended outcomes.^110,114

It should also be noted that the current findings refer only to studies using 1 to 20 Hz rTMS and conventional tDCS of dlPFC in ADHD. Other noninvasive brain stimulation protocols may be effective in ADHD, such as theta burst stimulation, ¹¹⁵ transcranial alternating current stimulation,^116,117 transcranial random noise stimulation^118,119 or trigeminal nerve stimulation (TNS).^120,121

Limitations

Conclusive evidence of this systematic review of rTMS and tDCS and meta-analysis of anodal tDCS in ADHD is limited by the large heterogeneity between studies with respect to stimulation protocols (coil/electrode placement, number of sessions, stimulation intensity, crossover/parallel design), sample age and cognitive outcome measures.

Furthermore, limitations in individual studies were also present, such as tDCS electrode placement, low power, small effect sizes, biased reporting or insufficient blinding. Specifically, although all but 2 of 12 tDCS studies included in the meta-analysis stimulated the left dlPFC, electrode placement was bilateral (anode left/cathode right) in 2 studies^66,68 and bifrontal (anode left and right) in 3 studies.^42,60,61 Bilateral or bifrontal tDCS could have had a neutral effect, given that the short inter-electrode distance causes greater current shunting across the scalp and cerebral spinal fluid, resulting in only an estimated 35% of current reaching the brain.¹²² In fact, it has been shown that bilateral electrode montage over the primary motor cortex can have neurologically neutral effects. ¹²³ With regards to low power, only 2 studies^65,66 had more than 30 participants; the sample size across all other studies was relatively small, with an average of 14 participants (with an n range of 7 to 20). Key outcome measures in interference inhibition tasks such as the Stroop and flanker reaction time or error interference effects (incongruent – congruent reaction times/errors) were not reported; instead, reaction times to incongruent trials were used to measure cognitive effects, meaning that effects on interference inhibition measures were confounded by processing speed.^64,67,68 Blinding integrity was not reported in 2 studies,^43,64 failed in 3 studies^65,67,71 and was potentially compromised in 3 studies, where at least 60% of participants correctly identified the stimulation type;^62,63,74 thus, placebo effects cannot be excluded. The meta-analysis of tDCS studies was hampered by the fact that only a minority of included studies controlled for baseline differences by analyzing change scores (i.e., post-treatment minus baseline;^62,65,66), while other studies analyzed only post-measurement scores to establish effects. Finally, the meta-regression analysis showed that studies with a high risk of “other” biases (e.g., no correction for multiple testing, selective reporting of outcome measures or using a lenient significance threshold^{43,62–64,67,68}) reported larger effect sizes than studies with low or unclear risk of this bias, meaning that summary effect size estimates might have been overestimated. In early rTMS studies, the same research practices unduly inflated positive effects and slowed its uptake as an effective treatment of depression; ^124,125 the field risks doing the same with rTMS and tDCS in ADHD. Moreover, with tDCS there is an added danger that children and parents — faced with apparent positive findings — will self-administer given the widely available “do-it-yourself tDCS” material online or commercial devices, one of which has been shown to impair working memory.⁴⁰ Given the neuroethical concerns of brain stimulation with respect to potential negative effects on nontargeted functions,^38,108,114 future researchers are duty-bound to report results to the highest possible standard.¹¹⁴

A final limitation of this meta-analysis is that it was not preregistered.

Conclusion

Based on current evidence, neither rTMS nor tDCS of the dlPFC can be recommended as an alternative neurotherapy for ADHD as yet. More studies are needed to assess clinical efficacy, and the demonstrated cognitive effects have been small and nonsignificant. However, we cannot rule out the possibility that rTMS or tDCS or other stimulation modalities of other regions — or even of the same region but using a larger number of sessions, different amplitude or other parameters, or combined with cognitive training — may be more effective. Furthermore, conclusive evidence from this systematic review of rTMS and tDCS studies and meta-analysis of tDCS studies in ADHD was hampered by heterogeneity in stimulation protocols, sample age and cognitive outcome measures. Larger, double-blind, randomized controlled trials with homogeneous protocols testing systematically for more optimal designs (e.g., multi-session stimulation combined with cognitive training, targeting right dorsal and ventral frontal or inferior parietal regions), ^9,126 and testing both clinical and cognitive outcomes, are needed to provide better insights into clinical and cognitive effects, and to provide clear guidance on optimal stimulation protocols. Future studies should also be wary of overstating positive effects and account for possible cognitive costs of tDCS and rTMS in children.