Evidence-Based Interventions for ADHD

ADHD is commonly treated with medication, but these treatments frequently cause side effects such as reduced appetite and disrupted sleep. Psychological and behavioral therapies exist as alternatives, but they tend to be expensive, hard to scale, and generally do little to address the motor difficulties that many children with ADHD experience — things like clumsy movement, poor handwriting, or difficulty with coordination. 

Physical exercise has attracted attention as a more accessible option. But research findings have been mixed, partly because studies vary so widely in how exercise is delivered and what outcomes they measure. This meta-analysis, drawing on 21 studies involving 850 children and adolescents aged 5–20 with a clinical ADHD diagnosis, tries to cut through that noise. 

Two types of motor skills 

The researchers separated motor skills into two broad categories: 

• Gross motor skills — movements involving large muscle groups, such as running, jumping, throwing, and maintaining balance 

• Fine motor skills — precise, controlled movements, typically of the hands and fingers, such as handwriting and manual dexterity (the ability to handle objects skillfully) 

‍

The Data: 

Gross motor skills (16 studies, 613 participants) 

Overall, exercise produced medium-to-large improvements in gross motor skills. The strongest gains were in: 

• Object control (e.g., throwing, kicking) — large improvement 

• Locomotion (e.g., running, swimming), body coordination, and strength — medium improvements 

No significant gains were found in balance or flexibility. 

Fine motor skills (13 studies, 553 participants):

Exercise also produced medium-to-large improvements in fine motor skills, specifically: 

• Handwriting: large improvement 

• Manual dexterity: medium-to-large improvement 

• Hand-eye coordination: moderate improvement 

The Results: What Kind of Exercise Works Best? 

Two factors stood out consistently across both gross and fine motor skills: session length and frequency. 

• Sessions longer than 45 minutes produced roughly twice the benefit of shorter sessions 

• Three or more sessions per week outperformed less frequent programs for gross motor gains 

The type of exercise mattered; structured programs with clear motor-skill components (rather than unstructured physical activity) yielded stronger results. 

These results are not without caveats, however. The authors urge caution in interpreting these findings. A few key limitations include: 

• Potential Publication Bias:  Studies showing positive results are more likely to be published, which can inflate apparent benefits. For gross motor skills, adjusting for this bias reduced the effect size from medium-to-large,  to medium. 

• Active vs. Passive Controls: When exercise was compared against doing nothing (a passive control), improvements looked significant. When compared against regular school activities (an active control), the gains were no longer statistically significant. This is a meaningful distinction: it suggests exercise may be beneficial, but not dramatically more so than simply being physically active in a structured school setting. 

• Medication status: Most participants were taking ADHD medication, so it’s unclear how well these findings apply to unmedicated children who might stand the most to benefit from structured exercise. 

• Study quality: Many studies lacked proper randomization, weakening confidence in the conclusions. 

The Bottom Line 

This meta-analysis provides tentative moderate evidence that structured physical exercise can meaningfully support motor skill development in children and adolescents with ADHD — particularly when sessions run longer than 45 minutes and occur at least three times a week. The benefits appear most robust for object control, locomotion, handwriting, and manual dexterity. 

That said, the evidence base still has real gaps. The authors call for better-designed, fully randomized controlled trials with consistent methods, standardized ways of measuring exercise intensity, and greater inclusion of children and adolescents who are not on medication — all of which would help clarify when, how, and for whom exercise works best. 

‍

Treatment guidelines for childhood ADHD recommend medications as the first-line treatment for most youth with ADHD. Still, concerns about side effects and long-term outcomes have increased interest in non-pharmacological approaches. Researchers at Saudi Arabian Armed Forces hospitals recently conducted a network meta-analysis comparing several interventions, including mindfulness-based therapy, cognitive behavioral therapy, behavioral parent training, neurofeedback, yoga, virtual reality programs, and digital working memory training. 

Although the authors aimed to “provide a rigorous methodological approach to combine evidence from multiple treatment comparisons,” the study illustrates several pitfalls that arise when network meta-analysis is applied to a thin and heterogeneous evidence base. 

What Network Meta-analysis Can and Cannot Do:

Network meta-analysis extends conventional meta-analysis by combining: 

• Direct comparisons (treatment A vs. treatment B tested in clinical trials), and 

• Indirect comparisons (A vs. B inferred through a common comparator such as placebo or usual care). 

When the evidence network is large and well-connected, this approach can provide useful estimates of comparative effectiveness among many treatments. 

This method is not always best, however, as many networks are sparse. This is especially true in areas such as complementary or behavioral therapies. In sparse networks, estimates rely heavily on indirect comparisons, and single studies can exert disproportionate influence over the results. 

Conventional meta-analysis focuses on heterogeneity, meaning differences in results across studies within the same comparison. 

Network meta-analysis must additionally evaluate consistency, whether the direct and indirect evidence agree. 

However, when comparisons are supported by only one or two studies and the network is weakly connected, statistical tests for heterogeneity and consistency have very little power. In practice, this means the analysis often cannot detect problems even if they are present. 

Sparse networks also make publication bias difficult to evaluate. This concern is particularly relevant in fields dominated by small trials and emerging therapies. 

Why Such Treatment Rankings Are Appealing, but Potentially Problematic:

Many network meta-analyses summarize results using SUCRA, which estimates the probability that each treatment ranks best. 

SUCRA, or Surface Under the Cumulative Ranking, is a key statistical metric in network meta-analyses. It is used to rank treatments by efficacy or safety. This is achieved by summarizing the probabilities of a treatment's rank into a single percentage, where a higher SUCRA value indicates a superior treatment. Ultimately, SUCRA helps pinpoint the most effective intervention among the ones compared. 

Again, in well-supported networks, SUCRA can provide a useful summary of comparative effectiveness. But in sparse networks, rankings can create an illusion of precision, because treatments supported by a single small study may appear highly ranked simply due to random variation. 

What Did this New Network Meta-analysis Study?

The study includes 16 trials with a total of 806 participants. But the structure of the evidence network is far weaker than this headline number suggests. 

Based on the underlying studies: 

• Six interventions are supported by a single trial each (digital cognitive mindfulness training, BrainFit, neurofeedback, online mindfulness-based program, cognitive behavioral therapy, and working-memory training) 

• Three interventions are supported by two trials each 

• Only one intervention is supported by three trials (family mindfulness-based therapy) 

This produces a very thin network, in which several interventions rely entirely on single studies. 

Another challenge is that the included trials measure different outcomes. Some evaluate ADHD symptom severity, while others measure parental stress. 

When studies use different outcome scales, meta-analysis typically relies on standardized measures such as the standardized mean difference to allow comparisons across studies. However, the analysis reports only mean-average differences, making it difficult to interpret the relative effect sizes. 

Study Issues (including Limited Evidence and Risk of Bias): 

The intervention supported by the largest number of studies (family mindfulness-based therapy) was one of the two approaches reported as producing statistically significant results. The other was BrainFit, which is supported by only a single previous trial. 

Despite this limited evidence base, the study ranks interventions using SUCRA: 

• Family MBT: 92% probability of being best 

• Behavioral parent training (BPT): 65% 

• Online mindfulness program: 49% 

• Cognitive behavioral therapy: 48% 

• Yoga: 39% 

Notably, none of the runner-up interventions demonstrated statistically significant efficacy. 

The authors acknowledge methodological limitations in the included studies: 

“Blinding of participants and personnel (performance bias) exhibited notable concerns, as blinding for active treatment was not applicable in most studies.” 

Such limitations are common in behavioral intervention trials, but they further increase uncertainty in already small evidence networks. 

Conclusions:

The study ultimately concludes: 

“This network meta-analysis supports MBT and BPT as effective non-pharmacological treatments for ADHD.” 

However, the evidence underlying these claims is limited. Some analyses rely on very small numbers of studies and participants, and the network structure depends heavily on indirect comparisons. 

Network meta-analysis can be a powerful tool when applied to a large, consistent, and well-connected body of evidence. When the evidence base is sparse, however, the resulting rankings and comparisons may appear statistically sophisticated while resting on a fragile evidentiary foundation.

‍

ADHD is one of the most common neurodevelopmental disorders in children, yet anyone familiar with this disorder, from clinicians and researchers to parents and patients, knows how differently it can manifest from one individual to the next. One person diagnosed with ADHD may primarily struggle with focus and staying on-task; another may find it nearly impossible to regulate their impulses or even start tasks; a third may frequently find themselves frozen with overwhelm and subject to emotional reactivity…

These are not just variations in severity; they may reflect genuinely different patterns of brain organization.

Our current diagnostic system groups all of these presentations under a single label (ADHD), with three behavioral subtypes (Hyperactive, Inattentive, and Combined) defined by symptom checklists. This framework has real clinical value of course, but it was built from behavioral observation rather than neurobiology, and may leave room for substantial heterogeneity to remain unexplained. In a new study, published in JAMA Psychiatry, researchers asked whether it’s possible to identify distinct neurobiologically subgroups within ADHD by analyzing patterns of brain structure, and whether those subgroups would map onto meaningful clinical differences.

How the Brain Was Analyzed

Researchers analyzed structural MRI scans from 446 children with ADHD and 708 typically-developing children across multiple research sites. From each scan, they constructed a morphometric similarity network; that is, a map of how different brain regions resemble one another in their structural properties. These networks reflect underlying biological organization, including shared patterns of cellular architecture and gene expression across brain regions.

From each individual's network, the research team calculated three properties that capture how each brain region functions within the broader network: how many connections it has, how efficiently it communicates with other regions, and how well it bridges different functional communities in the brain. Regions that score highly on these measures are sometimes called "hubs" and they play particularly influential roles in how information is integrated across the brain.

Rather than comparing the ADHD group to controls as a whole and looking for average differences, they used a normative modeling approach. This works similarly to a growth chart in pediatric medicine: instead of asking whether a child is above or below the group average, it asks how much a given child deviates from the expected range for their age and sex. This allows for individual variation across the ADHD group rather than flattening it into a single average profile.

The team then applied a data-driven clustering algorithm to these individual deviation profiles, allowing the data to reveal whether subgroups of children with ADHD shared similar patterns of brain network atypicality, without using any clinical symptom information to guide the clustering.

The Results:

Three stable, reproducible subtypes emerged from this analysis.

The first subtype was characterized by the most widespread differences from the normative range, particularly in regions connecting the medial prefrontal cortex to the pallidum (a deep brain structure involved in motivation and emotional regulation). Children in this group had the highest levels of both inattention and hyperactivity/impulsivity, and over a four-year follow-up period showed more persistent difficulties with emotional self-regulation than the other groups. They also had a higher rate of mood disorder comorbidity during follow-up, though this difference did not reach statistical significance given the sample size. The brain deviation patterns of this subtype showed correspondence with the spatial distributions of several neurotransmitter systems, including serotonin, dopamine, and acetylcholine, all of which have been previously implicated in ADHD pathophysiology.

The second subtype showed alterations concentrated in the anterior cingulate cortex and pallidum, a circuit involved in action control and response selection. This subtype had a predominantly hyperactive/impulsive profile, and its brain deviation patterns were associated with glutamate and cannabinoid receptor distributions.

The third subtype showed more focal differences in the superior frontal gyrus, a region involved in sustained attention. This subtype had a predominantly inattentive profile, with brain patterns linked to a specific serotonin receptor subtype.

A particularly important observation was that these brain-derived groupings aligned with clinically meaningful symptom differences, even though no symptom information was used in the clustering process. The fact that an analysis of brain structure alone arrived at groupings that correspond to recognizable clinical patterns is meaningful evidence that these subtypes reflect genuine neurobiological differences rather than statistical noise.

Replication in an Independent Sample

Scientific findings are only as trustworthy as their ability to replicate. The research team tested this clustering model in an entirely independent cohort of 554 children with ADHD from the Healthy Brain Network, a large, publicly available dataset collected under different conditions. The three subtypes were successfully identified in this new sample, with strong correlations between the brain deviation patterns observed in the original and validation cohorts. Differences in hyperactivity/impulsivity across subtypes were consistent with the discovery cohort, providing meaningful external validation of the approach.

What This Does and Doesn't Mean

It is important to be clear about what these findings do and do not imply. This study does not establish that these three subtypes are categorically distinct biological entities with sharp boundaries. They probably represent distinguishable regions along an underlying continuum of neurobiological variation. The neurochemical associations reported are exploratory and spatial in nature; they describe correspondences between brain deviation maps and neurotransmitter receptor density maps derived from separate imaging studies, and do not directly establish that any particular neurotransmitter system is altered in each subtype, nor do they currently inform treatment decisions.

The samples were not entirely medication-naive, and the strict comorbidity exclusion criteria may limit how well these findings generalize to typical clinical populations where comorbidities are the rule rather than the exception. All data came from research sites in the United States and China, and broader generalizability remains to be established.

What the study does demonstrate is that structured neurobiological heterogeneity exists within the ADHD diagnosis, that it can be reliably detected using brain imaging and data-driven methods, and that it aligns with meaningful clinical differences. The subtype defined by the most extensive brain network differences and the most severe, persistent clinical profile may be of particular importance, representing a group that could benefit most from early identification and targeted support.

The longer-term goal of this line of research is to move toward a more biologically grounded understanding of ADHD that complements existing diagnostic approaches and that may ultimately help guide more individualized treatment decisions. That goal, for now, remains a research ambition rather than a clinical reality, but this study takes a meaningful step in that direction.    

‍