Three ADHD
“biotypes” —
which the study
itself says may
be points on
a continuum

Three types of ADHD, again

In late September, National Geographic sent a February study back round the internet under a simple line: there might be three different types of ADHD. The post told its readers that new brain imaging research suggests ADHD “may fall into three distinct” biotypes, “offering a new explanation for why treatment can feel like trial and error.” The post drew more than 1,100 comments.

The study is real, large and carefully built. Nanfang Pan, Yajing Long, Qiyong Gong and colleagues — teams from West China Hospital, Monash University and the University of Cincinnati among them — published it in JAMA Psychiatry on 25 February 2026. It is worth reading in full, because the paper is more interesting, and more candid, than the story told about it.

What the study actually did

The discovery cohort was 446 children diagnosed with ADHD and 708 typically developing controls, drawn from six sites including the ADHD-200 datasets, with some control samples from the ABIDE initiative. From structural MRI the team built morphometric similarity networks — maps of how alike different grey-matter regions are to one another — and scored three properties of each region’s place in that network. Each child with ADHD was then compared with a normative model built from the controls, so that every brain region got a score for how far it sat from the typical range.

Those deviation scores were fed to a clustering method called HYDRA. It is described as semisupervised, and it is worth being exact about what that means here, because it is the obvious place to look for circularity: the only labels the algorithm uses are case versus control. It draws boundaries that separate children with ADHD from the control distribution, and each boundary becomes a candidate subgroup. No symptom scores and no DSM presentation labels went into the clustering. The authors say their clusters “converged with clinical phenotypes without using any clinical features,” and on the method as reported, that is fair.

The best solution was three groups: 142, 177 and 127 children. The team then looked at each group’s symptom scores and named them — severe-combined with emotional dysregulation, predominantly hyperactive/impulsive, and predominantly inattentive. A pretrained version of the model was applied to a second, independent cohort of 554 children with ADHD from the Healthy Brain Network, and the authors report that the core pattern replicated.

Where the names outrun the numbers

The brains came first, and the names came afterwards. That order matters, because the names are the part that travelled — and the names do more work than the data underneath them.

Start with how different the groups actually are in symptoms. Across the three biotypes, the differences in inattention and in hyperactivity/impulsivity were statistically significant but small: each accounted for roughly one and a half per cent of the variance between children (η² = 0.016 and 0.014). On the hyperactivity/impulsivity scale, rescaled from 0 to 1, the biotype named predominantly hyperactive/impulsive averaged 0.65; the biotype named predominantly inattentive averaged 0.60. On inattention the order flips, but only just: 0.75 for the inattentive biotype against 0.71 for the hyperactive/impulsive one. And the biotype that scored highest on hyperactivity/impulsivity was neither of them — it was biotype 1, at 0.68. These are overlapping distributions with slightly different centres, not three kinds of child.

Then look at the sample’s own diagnoses. The paper maps its biotypes against the children’s DSM presentations, and the figure is striking.

DSM presentation	Children	Brain-derived biotype	Children
Combined	246	Severe-combined with emotional dysregulation	142
Predominantly inattentive	191	Predominantly inattentive	127
Predominantly hyperactive/impulsive	9	Predominantly hyperactive/impulsive	177

Source: Pan et al. (2026), JAMA Psychiatry, Figure 4A and Results. Rows pair labels, not the same children.

Nine children in the sample carried a hyperactive/impulsive DSM presentation. The biotype given that name holds 177. Most of them, necessarily, were diagnosed as combined or inattentive. The label is a description of which way a group’s average leans on a symptom scale — not a claim that these children are hyperactive-impulsive in the clinical sense the phrase carries everywhere else, including in an ADHD assessment. Read by a parent on Instagram, it will not be taken that way.

The case–control brain differences underneath the whole exercise were small too. Children with ADHD showed more extreme deviations than controls on two of the three network measures, with effect sizes of d = 0.12 and 0.17; the third showed no significant difference. The authors themselves describe the overlap between ADHD and control brains as evidence of “the subtle and complex nature of ADHD-related brain alterations.” And the cluster stability index for the three-group solution was 0.21 — the best of the options tested, and still modest.

A paper that disagrees with itself

The most revealing thing in the paper is not a number. It is two sentences a few paragraphs apart.

Both cannot be the headline. The first is the one that sells, and it is the one the three-types framing carries. The second is the one the data supports: the symptom differences are slight, the brain differences are subtle, and the groups sit on continua that the clustering has sliced. The paper’s own conclusion sides with the first, calling the biotypes distinct, even though its methods and limitations call them putative.

The same gap runs through the claims about development and treatment. The conclusion credits the biotypes with distinct “longitudinal trajectories”. In the follow-up data, from medication-naive children at one site, there was no significant difference between biotypes over time in attention or externalising problems; the one difference found was in emotional self-regulation. Mood disorders that developed during follow-up were more common in biotype 1 (25.0%, against 9.8% and 5.0%), but the difference was not statistically significant. In the independent cohort, the hyperactivity pattern held but the inattention profiles “could not be fully reproduced”.

And on treatment — the hook of the Instagram post — the paper is explicit. The biotypes’ links to neurotransmitter receptor maps are called “exploratory”, and the authors write that these correspondences “cannot represent actual alterations in receptor systems themselves or inform treatment approaches.” The study does not test treatment at all. Its abstract says the biotypes “lay the groundwork for personalized management”; its main conclusion, more carefully, that the approach “may ultimately create a path toward developing personalized therapeutic strategies.” That is a hope about future studies, not a finding about why medication feels like trial and error. Two of the authors disclose research funding and consulting fees from pharmaceutical companies, reported in the paper as outside the submitted work.

Why the categories keep coming back

None of this makes the study bad science. It is a sophisticated attempt to ask whether ADHD’s variety has a structure in the brain, and its honest answer — found in its own limitations — is that the structure looks more like a continuum than a set of kinds. That is a useful result. The same question — kinds or continua — is live in the autism subtype research, and it surfaces more plainly in the transdiagnostic research, where the diagnostic categories fail to carve the data at clean joints.

What is worth noticing is what happens next. A data-driven method finds overlapping groups; the groups are named in the vocabulary of the existing manual; the names are then reported as the brain confirming the manual. The clustering did not use the DSM. The naming did. And once named, three points on a continuum become three types of ADHD, and the continuum drops out of the sentence.

That is the instrument encoding the verdict one step later than usual — not in how the data were collected, but in how they were described. ADHD’s categories have been drawn and redrawn by the DSM’s committees for decades; the construct survives partly because systems need it, not because it names a single thing. A brain study that finds a continuum and calls it three entities does not settle that argument. It restages it, with a scanner in the room. The more exact headline was in the paper all along: ADHD’s brain differences are subtle, they vary along a spectrum, and the three names are a way of talking about positions on it.

Citations

Pan, N., Long, Y., Qin, K. et al. (2026) — Mapping ADHD Heterogeneity and Biotypes by Topological Deviations in Morphometric Similarity Networks — JAMA Psychiatry