Study Design
We conducted a randomized, sham-controlled, double-blind trial of unmedicated children diagnosed with ADHD. The study CONSORT diagram is given in Figure S1. Twenty-five children were assessed for eligibility, 24 were randomized, and 23 participants completed the study. Only 1 participant was excluded from the study, due to difficulties complying with the required frequent arrival to the lab for treatment during the COVID-19 pandemic.
Study design is depicted in Fig. 1. All children were newly diagnosed and drug naïve. Following screening, eligible participants were assessed at baseline and then randomized into receiving either tRNS + CT or sham + CT for 2 weeks (weeks 1–2). Each group received their designated treatment for 5 consecutive days each week (one session each weekday). Participants were then assessed again with the same battery at the end of week 2 (t1), and 3 weeks later (t2), to examine endurance of effects. Parents and children, as well as study RAs, were blinded to treatment assignment. The total duration of subject participation in the study was 6 weeks. All study-related activities were conducted in a research lab at the Hebrew University of Jerusalem.
[Insert Fig. 1]
Study Population
Recruitment period was between December 2019 and December 2021. Participants (6–12 y/o) were recruited among children who were referred to the ADHD clinic by their paediatricians, general practitioners, teachers, psychologists, or parents. All participants gave verbal assent for participation and their parents provided written informed consent. All study procedures comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human patients were approved by approved by the Helsinki Committee (IRB) of the Hebrew University and Hadassah Medical Center (Jerusalem, Israel). The study is registered at ClinicalTrials.gov (identifier NCT03104972).
A power analysis showed that the n = 20 would allow to detect an effect with a direction hypothesis, given our previous results [28], with power = .8, and α = .05 and an effect size of Cohen’s d of 1.19. This estimated effect size is supported by the only available meta-analysis for tRNS [47].
The following inclusion criteria were applied: (1) age between 6–12 y/o; (2) score above the standard clinical cut-off score for ADHD symptoms on the ADHD DSM-5 scales; (3) meeting criteria for ADHD according to DSM-5, using the “gold standard” procedure as described by the American Academy of Paediatrics, which includes a semi-structured interview of the patient and parents by a specialist in paediatric neurology and child development, a neurological examination. Children were excluded from the study if they had one of the following: (1) a chronic neurological disease, epilepsy in the participant or in a first-degree relative, intellectual disability, other chronic conditions, chronic use of medications, or other primary psychiatric diagnosis (e.g., depression, anxiety, psychosis); (2) any Axis-1 disorders, assessed using the Kiddie-SADS-Lifetime Version, Hebrew version, which uses the DSM–5 criteria; (3) girls who began the age of puberty, based on a self- and part-report puberty questionnaire. The tool was translated to Hebrew by the study staff; (4) existence of epileptiform activity based on prospective resting-state electroencephalography performed at screening. EEG records were standardized and recorded with g.Recorder software (gTec, Schiedlberg, Austria), using a 64-channel wireless electroencephalography cap system (g.Nautilus) with gel-based electrodes.
Outcome Measures
Primary Outcome
The primary outcome measure was ADHD symptom severity, determined using the total score of the parent-reported ADHD-RS diagnostic questionnaire[1]. This scale is of well-accepted validity and reliability, regarded as standards in ADHD diagnosis and treatment effect. The scale contains 18 items based on the wording used to describe those items in the DSM-5: the first 9 items measure inattention (IN) symptoms, while the followed 9 items measure hyperactive-impulsive (HI) symptoms (see full description in [28]).
Secondary outcomes
Global functioning was measured using the CGI-S (Clinical Global Impression–Severity) scale[48], memory performance was measured using the Digit Span test [49] and PS was measured using the MOXO-CPT task (NeuroTech Solutions Ltd); These measures have been detailed in our previous publications (see [28], [29]). Everyday EFs were assessed using the Behaviour Rating Inventory of Executive Function (BRIEF, [50]), parent and teacher reports.
Sleep quality was assessed using the Hebrew version of the Pittsburgh Sleep Quality Index [51], a self-report questionnaire used to assess sleep quality and disturbances, designed for adults (see Supplemental Material 1). Here, children were asked to answer this questionnaire together with their parents, and one item was adapted to fit children’s daily life (the item ‘use of sleeping medication’ was replaced by ‘how many times you wake up at a night’).
RS-EEG. The full details on the EEG recording and pre-processing and given in Supplementary Material 1. In short, electrophysiological data was recorded using an eyes open (EO) resting condition. A fast Fourier transform (FFT) was used to calculate the absolute power spectra within different specific frequency bands, focusing on delta (.5 − 2 Hz), theta (34-7Hz), alpha (8–13 Hz), beta (13–30 Hz) and total power (1-40Hz) of all band changes in each group. Here, we focused on analysing the data from electrodes over the stimulation sites (F3, F8) as well as from frontal midline area (Fz), which has been shown changes in aperiodic exponent following tRNS applied to similar brain regions[46]. FFT has been extracted for each electrode.
We then employed a spectral parameterization approach which enables decomposition of the neural signal into its respective periodic and aperiodic components, with a tool called FOOOF (fitting oscillations and one over f) [41]. Importantly, the FOOOF tool calculates both the aperiodic value for each electrode and models the distribution features of the periodic component in the bands of interest. This also gives the central frequency and bandwidth of the periodic component's distribution. We note that we were not able to detect the peak for alpha and theta frequencies for most participants, and therefore did not have enough data to draw reliable conclusion for these frequencies. We therefore do not report results for these frequencies.
Study Interventions
A detailed description of the study interventions is given elsewhere ([28], [29]). In short, participants completed computerized CT along with either tRNS (tRNS + CT arm) or sham (sham + CT arm) for 20 minutes / day for 10 days during a 2-week period. Sessions were conducted daily each week from Sunday through Thursday, and no sessions were conducted on the weekend (Friday and Saturday).
For sham tRNS we used the same montage as in active tRNS, but here the 30 sec of ramp up of the current from 0 to 0.75mA was immediately followed by 30 sec ramp down period to 0mA, such that participants did not receive active stimulation between ramp-up and down. This method has been shown to provide effective blindness of the stimulation condition as both active and sham tES would lead to slight itching sensation that would disappear due scalp habitation [52].
Randomization And Blinding
Participants were randomized in a 1:1 allocation ratio to receive active or sham tRNS. Three staff members with no contact with participants generated balanced random samples throughout the course of an experiment, using Smith’s randomization algorithm based on the variance minimization procedure [53], and programmed the device to discharge sham/active stimulation according to each participant’s allocation. The active and sham tRNS were physically indistinguishable based on the electrode locations and the displayed information for RAs and participants. Participants and their parents were blind to the treatment assignment, as well as the PI and study staff involved in training and/or assessment. To examine the success of the blinding procedure, parents of participating children were asked at the end of the experiment which intervention they think that their child received and to rate the level of confidence in their prediction. The blinding assessment was performed using the Bang Blinding Index, ranging from − 1 to 1, with 1 indicating total lack of blinding, 0 indicating complete blinding and − 1 indicating opposite guessing which may be related unblinding [54]. A positive value suggests that parents correctly guessed their child’s treatment allocation beyond chance.
Statistical Analysis
All statistical analyses were conducted using R. Study staff who conducted the analyses were blind to group assignment during pre-processing and analysis of all measures. Overall, there was less than 4% missing data in the entire dataset, which stem from missing data in the scales of BRIEF teachers and RS-EEG recordings, as well as missing daily treatment sessions. Before statistical testing, outlier data, defined a-priori as values 2.5 SDs above or below the group mean of each measure, were removed from further analyses. The range of outliers across variables did not exceed 3% in the behavioral outcomes and were solely in the PS and sleep index scales, while in the RS-EEG recordings it did not exceed 6% and was just in F3 electrode. This was due to movement restrictions imposed during the COVID-19 pandemic, which affected arrivals to the lab and to schools. There were no significant group differences in terms of missing data, nor in outlier variables in all time points (p > 0.5).
Demographic characteristics of age, gender, and estimated IQ (WISC subscales) were compared using Student’s t-tests and chi-square tests for independent samples. ADHD symptoms were compared between groups using separate one-way MANOVAs, using the IBM SPSS Statistics version 25 (IBM Corp., Armonk, N.Y, USA). Linear mixed effects models (LMMs) were used to examine treatment effects. LMMs account for within-subject correlations and for associations induced by repeated measurements. To conduct LMM analyses, we used the R-package nlme with maximized log-likelihood on the outcome measures, and subjects as the random factor. We examined outcomes immediately post-treatment (t1) and at a 3-week follow-up (t2) for each condition and included treatment arm (tRNS + CT, Sham + CT) and time (t1 and t2) as predictors. Baseline performance was added as a covariate to the model, allowing for better adjustment for minor differences in the pre-treatment means.
For our primary outcome measure (ADHD-RS), a simple model which included the main effects of stimulation and time with no interaction between them was preferred to a more complex model that included the interaction term (F(7) = 2.06, p = .15; see also [28]). We therefore report this parsimonious model for the secondary measures as well. For all measures, we verified that the residuals were normally distributed using a q-q plot and the Shapiro-Wilk normality test. The only exceptions were the SOL index residuals, and some of RS bands (rIFG: delta, alpha and beta; lDLPFC: delta and alpha) which were not normally distributed; we therefore applied log10 transformations to normalize these measures.