Eligible participants were Dutch speaking children, 7–13 years of age, with a primary clinical DSM-IV-TR diagnosis of ADHD . Children with ADHD were recruited from fifteen child mental health outpatient care facilities in the West of the Netherlands. Before entering the study, parent- and teacher ratings on the Disruptive Behavior Disorders Rating Scale (DBDRS)  confirmed their diagnosis; at least one of the scores on the Inattention or Hyperactivity/Impulsivity scales had to be above the 90th percentile for one of the informants, and above the 70th percentile for the other informant (signifying pervasiveness of symptoms). At study entry, all children were free of stimulant use for at least 1 month. Exclusion criteria were neurological disorders and IQ below 80 as measured by a four subtest version of the Wechsler Intelligence Scale of Children-III (WISC-III) including the subtests Vocabulary, Arithmetic, Block Design, and Picture Arrangement . No restrictions were set on other comorbidities. Comorbid disorders were diagnosed according to DSM-IV-TR and retrieved from the medical records. Comorbid disorders included learning disorders (NFB; n = 5, MPH; n = 2, PA; n = 1), autism spectrum disorders, (NFB; n = 3, MPH; n = 2, PA; n = 3), anxiety disorders (NFB; n = 2, MPH; n = 0, PA; n = 2), and mood disorder (NFB; n = 1, MPH; n = 0, PA; n = 0). Chi-square test revealed no significant difference in the distribution of comorbid disorders over groups, χ
2 (8, N = 112) = 12.88, p = 0.12.
Initially, 112 children with ADHD were randomized over the three interventions (NFB; n = 39; MPH; n = 36; PA; n = 37), with 103 children completing their intervention (NFB; n = 38; MPH; n = 31; PA; n = 34). Drop-out reasons included motivational and/or practical reasons (NFB; n = 1, MPH; n = 3, PA; n = 3) and medical contraindications (MPH; n = 2). A participant flow diagram is presented elsewhere .
A multicentre three-way parallel group study with balanced randomization was conducted. A randomization table was created using a computerized random number generator . Stocks of nine unmarked sealed envelopes were presented to parents at intake. Parents randomly picked an envelope revealing intervention allocation. Subsequently, children, parents, and teachers were aware of the allocated group. Data collection took place between September 2010 and March 2014.
The current study aimed to enroll 186 participants. In total, 135 children with ADHD were assessed for eligibility and eventually 112 participants were randomized over the three interventions. To detect a medium effect size (f = 0.25) using three groups in a repeated measures (RM) analysis of variance (ANOVA) with an alpha 0.05 and a power of 95 %, a total sample size of 66 (i.e. 22 per group) was required . Post-hoc comparisons of two groups required a total sample size of 54 (i.e. 27 per group) to detect a medium effect size (f = 0.25) in a RM ANOVA with an alpha 0.05 and a power of 95 %. In the current study, the smallest group size was 29. Consequently, all groups had enough participants to detect a medium effect size. Because in total 112 participants were randomized over the three groups instead of 186 participants, the current study did not achieve the statistical power needed to detect smaller effect sizes than medium (f < 0.25) between groups. This report complies with the CONSORT 2010 guidelines (Supplement Appendix 1) for reporting parallel group randomized trials . The trial was registered on clinicaltrials.gov (Ref. No. NCT01363544).
NFB and PA treatment consisted of three individual training sessions a week, with each session lasting 45 min including 20 min of effective training, over a period of 10–12 weeks. All interventions, as described below, took place after the pre-intervention (t0) assessment.
Neurofeedback (NFB). Theta/beta training was applied with the aim to inhibit theta (4–8 Hz) and reinforce beta (13–20 Hz) activity at Cz. Theta/beta index was represented to the participant by simple graphics on a screen. Successful reduction of the theta/beta index as averaged over one trial relative to session baseline, was rewarded with the appearance of a sun and granted with credits. To promote generalization of the learned strategies into daily life, transfer trials were used. Transfer trials were presented without immediate visual feedback and were included from session 11 (25 %) and session 21 (50 %) onwards. To further transfer learned behaviors, participants were instructed to retrieve their neurofeedback experiences by watching printed graphics of the training during school and homework. Compliance was verified by questioning the participants whether they used the transfer cards over the intervention period. Transfer cards were used by 84 % of the participants. See also Supplement Appendix 2 for more detailed information about the neurofeedback intervention. The mean number of training sessions of participants who completed the assessments at post intervention (n = 38) was 29 (M = 28.53, SD = 2.63, range between 19 and 30).
Medication (MPH). After the pre-intervention assessment, a 4-week double-blind randomized placebo-controlled titration procedure was used to determine the optimal individual dose of short-acting methylphenidate (MPH) . The 4-week titration phase was preceded by a baseline week to determine ADHD symptoms without MPH, and was followed by a lead-in week in which on three consecutive days, twice-daily (at breakfast and lunch time), doses of (1) 5 mg, (2) 10 mg, and (3) 15 mg (<25 kg body weight) or 20 mg MPH (>25 kg body weight) were used to assess possible adverse effects. During the 4 weeks titration phase, children received in pseudo-random order (1) 5 mg, (2) 10 mg, (3) 15 mg or 20 mg MPH or (4) placebo for 1 week, twice daily. During the titration phase, children, parents and teacher as well as the researchers were blind with regard to the prescribed dose. At the end of each week, parents and teacher were asked to evaluate inattention and hyperactivity/impulsivity symptoms on the DBDRS, and adverse effects on the MTA Side Effects Rating Scale . In total, 31 children completed the titration procedure. Children were classified by a standardized procedure  as responders when their ADHD symptoms significantly decreased compared to placebo (n = 29). The standardized procedure  classified children as non-responders when they did not show any decrease in inattention and hyperactivity/impulsivity symptoms across MPH doses and placebo as compared to baseline assessments (n = 2). When children were found to respond equally well across different MPH doses, the lowest MPH dose was prescribed. The two non-responders were treated with 5 mg MPH twice daily. The child’s psychiatrist prescribed the optimal dose for the remaining intervention period (5 mg to 10 children including 8 responders and 2 non-responders, 10 mg to 14 children, 15 mg to 2 children, and 20 mg to 5 children).
Physical activity (PA) as semi-active control condition. Maximum heart rate (HRmax) was determined before the start of the first training session. Each training session started with 5 min of warming up, followed by five 2-min moderate intensity exercises at a level of 70–80 % of HRmax. After a 5-min break, five 2-min vigorous intensity exercises of 80–100 % of HRmax were performed. Each training finished with a 5-min cool down. Time and heart rate were monitored and registered using a POLAR FT4 watch (Polar Electro Oy, Kempele, Finland). The mean number of sessions of participants who completed the assessments at post intervention (n = 34) was 28 (M = 27.74, SD = 3.56, range 12–30).
The auditory oddball task was used to measure attention . This task contained 255 standard tones (523 Hz, 85 %) and 45 target tones (1046 Hz, 15 %), presented pseudo-randomly for 100 ms. Children were instructed to attend to the stimuli and to press a button on a response box with the right index finger when they heard a target. Outcome measures were response speed (mean reaction time; MRT), assessing attention, and the coefficient of variation (CV) [CV = MRT SD/MRT], a measure of attentional lapses . Omission and commission errors were uncommon and, therefore, excluded from analyses.
The stop-signal task (SST) was primarily used to measure inhibition . This task required children to perform a binary-choice reaction time task using visual stimuli (go stimuli). Children were instructed to inhibit their response when a go stimulus was followed by a visual stop signal. A full description of the task can be found in Janssen et al. . Variables of interest were: (1) stop-signal reaction time (SSRT), a measure of the speed of the inhibitory process, calculated by subtracting mean stop-signal delay (SSD) from MRT; (2) number of commission on go trials, measuring impulsivity; (3) number of omission errors on go trials, assessing attention; (4) response speed (MRT), and (5) variability of response speed as calculated by coefficient of variation (CV), measuring lapses of attention.
The visual spatial working memory task (VSWM) [45, 46] was assessed to measure short-term storage or maintenance of visual-spatial information (forward condition) and visuospatial working memory (backward condition). Children were instructed to repeat sequences of yellow circles, presented on a computer screen in a 4 × 4 grid, in a forward order (forward condition) and a reversed order (backward condition). Variables of interest were the number of correct trials per condition.
The study was approved by the national medical ethics committee (NL 31641.029.10 CCMO). Written informed consent was obtained before participation from all parents and children aged 11 and older.
Pre-intervention (t0) assessment took place in the week prior to the start of the intervention. Post-intervention (t1) assessment took place 1 week after the last training session. Part of the data of this study are presented elsewhere [34, 47]. During t1 assessment, the MPH-group continued use of medication. Due to technical problems or misinterpretation of the task, data of 23 participants for the oddball task and 10 participants for the stop-signal task were not available for analysis. Finally, data of 89 participants for the oddball task (NFB n = 30; MPH n = 29; PA n = 30) and 102 participants for the stop signal task (NFB n = 36; MPH n = 33; PA n = 33) were analyzed. Interventions took place between September 2010 and March 2014.
Statistical analyses were performed with the IBM SPSS Statistics, version 20.0 . Differences between treatment groups in terms of background characteristics were analyzed with a Chi-square test or ANOVA with Tukey post hoc tests. Group characteristics and outcome measures were subjected to attrition analyses using ANOVA, comparing the initially randomized sample to the sample that completed the interventions.
To compare treatment effects, General Linear Model (GLM) repeated measures (RM) ANOVAs were applied, with time [between pre-intervention (t0) and post-intervention (t1)] as within-subject factor and group (NFB, MPH and PA) as between-subject factor. For these analyses, the adjusted difference at post-intervention [ADt1-t0] and accompanying 95 % confidence interval (95 % CI) and the accompanying effect size (partial eta squared, η
) are reported. Effect sizes are expressed in percentage of explained variance in partial eta squared (η
; with thresholds for small, medium, and large effects corresponding to η
= 0.01, η
= 0.06, and η
= 0.14, respectively . In case of significant time by group interactions, post hoc two-way between-groups interactions analyses were performed separately for the between-subject factors (1) NFB and MPH, (2) MPH and PA and (3) NFB and PA with time (t0, t1) as within-subject factor. Only significant results of p ≤ 0.05 are reported. Intention-to-treat analyses were performed using imputation with Last Observation Carried Forward (LOCF). Complete case analyses were performed for participants who completed pre- and post-intervention assessments. Post hoc analyses were performed, with separate addition of assessment site (Amsterdam or Rotterdam) and comorbid disorders (ADHD or ADHD with comorbid disorders). At group level, we found participants in the neurofeedback training were getting better at decreasing theta/beta ratio over time . Therefore, to investigate the relation between getting better at decreasing theta/beta ratio (EEG slopes) and improvement in cognitive measures (difference scores), Pearson correlations were computed. Decreased theta/beta ratio was represented by (1) theta slopes over runs (within sessions) and theta slopes over sessions and (2) beta slopes over runs and theta slopes over sessions. Improvement in cognitive measures was defined by difference scores (t1-t0) of the variables of interest as described above.