The topic this week is ADHD
The articles
Please click on the arrow to show the details for each article.
Chen, Y. et al. (2025). Distinct neuroimaging subtypes of ADHD among adolescents based on semi-supervised learning. Translational Psychiatry. https://doi.org/10.1038/s41398-025-03662-3 Pan, N. et al. (2026). Mapping ADHD Heterogeneity and Biotypes by Topological Deviations in Morphometric Similarity Networks. JAMA Psychiatry. https://doi.org/10.1001/jamapsychiatry.2026.0001
Both studies are tackling the same underlying problem: ADHD, as we currently diagnose it, is a single label covering an enormously varied group of children. Two children can both meet criteria for combined-type ADHD and yet look completely different in the classroom, respond completely differently to medication, and have completely different co-occurring difficulties. Both research teams wanted to know whether brain imaging could explain some of that variation, essentially asking whether there are biologically distinct “subtypes” or “biotypes” hiding underneath the single diagnostic label.
Chen and colleagues used data from the Adolescent Brain Cognitive Development, or ABCD, study, looking at cortical thickness, that’s the thickness of the grey matter across the brain’s surface, in adolescents with and without ADHD. Their question was whether adolescents with ADHD cluster into distinct groups based on patterns of cortical thickness, and whether those groups differ in their social, clinical, and treatment characteristics.
Pan and colleagues took a related but more sophisticated approach. Rather than looking at the thickness of individual brain regions, they examined morphometric similarity networks, essentially how similar different brain regions are to one another in their structure, and measured how “hub-like” or well-connected each region was. Their question was whether atypical patterns in this brain-wide organisation could identify robust, clinically meaningful ADHD biotypes, and whether these biotypes had distinct neurochemical and clinical profiles.
Methods
I won’t dwell too long on the technical details here, as both use fairly complex machine learning methods, but it’s worth knowing the broad approach because it affects how much confidence we can place in the findings.
Chen’s team used a semi-supervised deep learning method called Smile-GAN to identify subtypes from cortical thickness data in 929 adolescents with ADHD and 5,580 typically developing controls, all drawn from the ABCD study. Crucially, they didn’t stop there. They checked whether the same subtypes appeared again two years later in the same participants. Separately they validated the subtypes in an entirely separate dataset, ADHD-200, with a further 330 adolescents with ADHD. They then linked the subtypes to socioeconomic and cognitive data, medication response, and underlying neurotransmitter systems.
Pan’s team used a different clustering method, called HYDRA, and instead of looking at cortical thickness, they considered the morphology and connectedness of different brain regions. Their discovery sample included 446 children with ADHD and 708 typically developing controls across five sites in the US and China, with age-and-sex normative models used to work out how far each child’s brain organisation deviated from what would be expected for their age. They then validated their biotypes in an independent sample of 554 children with ADHD from a different, transdiagnostic cohort, and linked the biotypes to symptom profiles, neurotransmitter maps, and cognitive processes.
Both teams had large samples and went to considerable lengths to check their subtypes weren’t a statistical fluke, validating them in independent samples using different scanners, sites, and even countries.
The Results
Three subtypes, twice over
Here’s the headline finding that makes reading these two studies together so interesting: using two different brain measures and two different clustering methods, in two different, non-overlapping samples of children and adolescents with ADHD, both teams landed on three subtypes.
Chen’s team identified
- an under-developed subtype, with thinner cortex particularly in posterior brain regions,
- an over-developed subtype, with thicker cortex in posterior regions,
- and a mixed subtype, with a scattered pattern of thicker and thinner regions.
Pan’s team, working with entirely different brain metrics, identified
- a severe-combined subtype with widespread alterations centred on the medial prefrontal cortex and pallidum,
- a predominantly hyperactive/impulsive subtype centred on the anterior cingulate cortex and pallidum,
- and a predominantly inattentive subtype centred on the superior frontal gyrus.
A subtype linked to disadvantage and stimulant response
In Chen’s study, the under-developed subtype stood out for its association with socioeconomic stress: these children had significantly lower family income, lower parental education, and younger maternal age at birth than the other subtypes, alongside the lowest cognitive scores. Yet, encouragingly, this was also the subtype that responded best to stimulant medication.
The over-developed subtype told the opposite story. These children didn’t show the same socioeconomic association, but they showed the poorest response to stimulant medication, particularly for hyperactive/impulsive symptoms, with little change in symptom severity despite treatment. The authors suggested this subtype has a genuinely different underlying biology, more strongly linked to insulin-related pathways than the dopamine pathways stimulant medication targets, which may explain the poor response.
A subtype linked to severity and emotional dysregulation
Pan’s severe-combined subtype, which they described as showing “emotional dysregulation,” was the most severely affected group, scoring highest on both inattention and hyperactivity/impulsivity measures, with the most widespread pattern of brain alterations of the three subtypes. It also showed a numerically, though not statistically significant, higher rate of co-occurring mood disorder, 25% compared with 5 to 9% in the other two subtypes. This subtype’s deviations correlated most strongly with the serotonin system, alongside dopamine, acetylcholine, and histamine receptors, consistent with a broader disruption across several neurotransmitter systems rather than a single pathway.
Different mechanisms behind inattention and hyperactivity
Both papers point towards inattentive and hyperactive/impulsive presentations having at least partially separate neural underpinnings, rather than simply sitting at different points on a single severity scale.
Chen’s team similarly found that the subtypes responding best to stimulant medication were those with structural differences in the dorsolateral prefrontal cortex, a region heavily implicated in attention and executive control. While the poorly responding over-developed subtype showed no such difference from typically developing peers in this region.
In Pan’s study, the predominantly hyperactive/impulsive subtype was centred on circuitry involved in action-selection networks. While the predominantly inattentive subtype was centred on circuitry linked to sustained attention and default mode network interference, essentially, the brain’s tendency to drift into internally focused thought at the expense of task focus.
Practical Takeaways
So, what does all this brain-scanning and clustering mean for those of us working with children in mainstream schools, where none of us will ever be ordering an MRI?
First, and most importantly, this is not something we can, or should, translate directly into practice yet. Neuroimaging subtyping remains a research tool. No school-based OT is going to receive a “biotype” alongside a diagnosis, and both research teams are explicit that further validation is needed before any of this informs clinical decision-making.
That underlying message is this: “ADHD” is not one thing. Reading these two studies together strengthens something we probably already sensed from clinical experience, that two children with the same diagnostic label can have meaningfully different presentations, different responses to intervention, and potentially different underlying difficulties. This should reinforce individualised, functional assessment over assuming a “standard” ADHD profile or applying a one-size-fits-all intervention approach.
Second, both studies found that inattentive and hyperactive/impulsive difficulties may reflect at least partly separate underlying mechanisms, rather than the same problem at different intensities. Clinically, this supports what many of us already do: tailoring strategies to the specific profile in front of us. A child whose difficulties are predominantly about sustained attention and mind-wandering may benefit from a different set of environmental and self-regulation strategies than a child whose difficulties are predominantly about impulsive action and movement, even where both children carry the same diagnosis.
Third, Chen’s finding around socioeconomic stress is a useful reminder, not that disadvantage causes “worse” ADHD, but that a child’s wider environmental context is intertwined with their presentation and, potentially, their profile of difficulty. This supports continuing to take a holistic, family-context-sensitive approach to assessment, and being cautious about attributing everything we see to a fixed, purely neurological explanation.
Fourth, if you work closely with prescribers, or families raise questions about medication, both studies offer a useful, evidence-based reminder that response to stimulant medication genuinely varies for biological reasons, not simply because of adherence or dosing. I think the studies provide evidence to support what we see clinically: not all children respond to medication. If a child isn’t responding well to stimulant medication, this doesn’t necessarily mean something is being done wrong; it may reflect a different underlying profile that needs a different approach altogether.
Finally, Pan’s finding that the most severe, most widely affected subtype also showed a numerically higher rate of co-occurring mood difficulties is worth holding onto. Where a child’s ADHD presentation involves marked emotional dysregulation alongside inattention and hyperactivity, this may signal a need for closer collaboration with mental health colleagues and earlier, more intensive support, rather than waiting to see whether difficulties resolve with standard intervention alone.
Limitations
Both studies have limitations worth flagging. Chen’s ABCD sample was predominantly of European ancestry, so findings may not generalise across other populations, and the associations with medication response, while promising, were based on relatively small subgroup sizes. Pan’s discovery and validation cohorts, while genuinely independent, drew on similarly resourced settings across the US and China, and the authors themselves caution that these “biotypes” may represent points along continuous dimensions of brain variation, rather than truly distinct categories. Neither study can tell us anything about causation; we don’t know whether these brain differences precede a child’s experiences, result from them, or some combination of both. And critically, neither study involved any occupational or functional outcome measures relevant to school participation, so translating these findings into classroom-relevant recommendations requires real caution.
Taken together though, these two studies, using different techniques, different samples, and completely independent research teams, arrive at a strikingly similar conclusion: there is real, replicable neurobiological heterogeneity hiding underneath the single label of ADHD. That’s a finding worth knowing, even while we wait for research to catch up with something we can put directly into practice.
OT Sensory Series
If you are an OT who wants to learn more about assessing sensory needs and using sensory based interventions in schools and home, make sure you check out my sensory series for OTs. The series is now available on demand, and you can watch in your own time. It’s perfect CPD for those working in schools or with families.
Join the mailing list
If you’re enjoying the podcast don’t forget to subscribe to the mailing list to be the first to hear when the next episode is released.
You can also book onto the live journal clubs here.
Disclaimer
This podcast provides educational commentary and analysis of recent research for continuing professional development. All studies are properly cited and used under fair use provisions for educational purposes. Listeners should consult original sources, using the links above, for complete study details.
