Skip to content
The Neurodiversity Directory

8,780 young
people, four
diagnoses,
no exclusive
signature — and
some ADHD
findings faded
once medication
was counted

Measured in one analysis, ADHD, conduct disorder, anxiety and depression showed small, mostly shared brain differences and no exclusive signature. For ADHD, part of the picture could not be separated from medication.

Four diagnoses, one analysis, and the question single-disorder studies could not ask

Most brain-imaging research on ADHD compares people with one diagnosis against people with none. That design can show a group differs from controls. It cannot show whether the difference belongs to that diagnosis, or whether youths with other diagnoses would show the same thing if measured the same way.

A mega-analysis from four working groups of the ENIGMA Consortium, led from the Department of Psychology at the University of Bath and published in the 15 August 2026 issue of Biological Psychiatry, measured them the same way. Sophie Townend, Graeme Fairchild and colleagues pooled structural MRI data from 8,780 young people aged 4 to 21: 1,317 with ADHD, 1,172 with conduct disorder, 1,044 with anxiety disorders, 504 with depression and 4,743 healthy controls. The 4,037 youths with a diagnosis and the 4,743 controls were analysed as individual participant-level data in a single cross-sectional mega-analysis, adjusted for site, age and sex.

Who was scanned

1,317
youths with ADHDTownend et al., 2026
1,172
youths with conduct disorderTownend et al., 2026
1,548
youths with anxiety disorders (1,044) or depression (504)Townend et al., 2026
4,743
healthy controlsTownend et al., 2026

The two internalising groups together number 1,548. The authors measured cortical thickness and surface area across 34 regions, and the volume of 7 subcortical structures.

What was shared, what looked specific, and why “specific” did not mean exclusive

Start with what all four groups had in common. Compared with controls, youths with ADHD, conduct disorder, anxiety and depression all showed lower surface area in the insula, entorhinal cortex and middle temporal gyrus, and lower amygdala volume, with effect sizes (Cohen’s d) from −0.07 to −0.24. All four also showed lower total surface area and lower intracranial volume. The authors read the insula and amygdala findings through the salience network, and suggest it “may be implicated in general psychopathological processes shared across diagnostic categories”.

ADHD and conduct disorder also shared lower surface area in frontoparietal regions that the internalising groups did not show (d from −0.08 to −0.13). No pattern was found that belonged to anxiety and depression alone. Conduct disorder showed the most widespread differences, with lower surface area in 21 of 34 cortical regions, and reductions in hippocampus, nucleus accumbens and thalamus volume that appeared in that group only.

For ADHD, the study reports three disorder-specific associations: lower surface area in the posterior cingulate cortex, and lower cortical thickness in the fusiform gyrus and temporal pole. Here the definitions matter. In this paper, “disorder-specific” means that one diagnostic group differed significantly from controls and the others did not. It does not mean the group also differed from the other three diagnoses, and the authors note cases where two groups showed similar effect sizes but only one reached significance, possibly because of differing sample sizes. On true uniqueness they are brief: “no examples of true exclusivity were identified”.

Their conclusion is stated at its own strength. The lack of exclusive effects and the extent of shared alterations “suggest that these diagnostic groups may have more in common than was previously thought.” They go one step further on cortical thickness, which showed fewer and smaller differences than surface area: their results suggest that many earlier single-disorder findings of altered cortical thickness in youth “might have been false positives”, which may explain why those findings have been hard to replicate.

The condition or the prescription: what the medication adjustment can and cannot tell us

The finding that should travel furthest sits in the sensitivity analyses. The authors re-ran their models adjusting for IQ, and separately for whether each young person was taking psychotropic medication at the time of the scan. Most cortical results survived the IQ adjustment and were more affected by medication; for subcortical volume the reverse held. Ninety per cent of the disorder-specific associations survived adjustment for IQ, “while only 42% survived adjustment for medication status”. That 42% covers the disorder-specific associations across all the diagnoses, not ADHD alone.

For ADHD, the one case the paper spells out in its main text is the frontoparietal pattern it shared with conduct disorder. After the medication adjustment, only the conduct disorder group still differed from controls in those regions: in the paper’s words, “ADHD-related effects were nonsignificant”. The transdiagnostic amygdala reduction, by contrast, survived the medication adjustment but not the IQ adjustment.

The reason ADHD is most exposed is visible in the paper’s own sample table. Among participants with medication data, 41.96% of the ADHD group (base 908) were taking psychotropic medication when scanned, against 23.05% of the conduct disorder group (base 1,041), 14.25% of the depression group (base 372), 10.93% of the anxiety group (base 979) and 1.42% of controls (base 4,158).

Taking psychotropic medication at the time of the scan

Share of each group, among participants with medication data; bases: ADHD 908, conduct disorder 1,041, depression 372, anxiety 979, controls 4,158

  1. ADHD41.96%
  2. Conduct disorder23.05%
  3. Depression14.25%
  4. Anxiety disorders10.93%
  5. Healthy controls1.42%

Source: Townend et al. (2026), Table 1, Biological Psychiatry

So is a structural difference in an ADHD brain the condition, or the prescription? The paper offers two readings and does not choose between them. One is that the adjustments suggest “some structural alterations were more related to medication use than to disorder status.” The other is that medication status may be a proxy for severity: children with severe ADHD are more likely to be prescribed psychostimulants, so adjusting for medication may partly be adjusting away the most severe cases. “Missing information on disorder severity prevented us from exploring this aspect here.”

Two further limits apply. Medication entered the analysis only as yes or no, for any psychotropic drug, at the time of the scan: nothing about which drug, what dose, or for how long. And medication data were missing for many sites and participants, so the authors ask for the sensitivity analyses to be “interpreted cautiously”. Prescription is not a pure readout of a child’s condition either. The Directory has reported a Norwegian registry study in which whether a child at the margin of treatment was medicated depended measurably on which clinic they attended. A variable shaped by services as well as severity is hard to adjust for cleanly.

At the paper’s own strength, then: part of what would be called the ADHD brain, in this sample, cannot be separated from the fact that more than four in ten of the ADHD group with medication data were medicated. Whether the medication produced those differences, or simply marked the children with the most severe presentations, is not known.

A third of a standard deviation is not a diagnostic test

The largest difference anywhere in the study had a Cohen’s d of −0.33. Surface area and subcortical effects ranged from −0.33 to 0.11. Cortical thickness effects ranged from −0.14 to 0.09. A difference of a third of a standard deviation between group averages leaves the two groups overlapping for most of their range. On differences this size, most young people with any of these diagnoses would have brain measures inside the range of the controls.

The authors suggest that differences within each diagnosis may have helped keep the effects small: anxiety disorders here combine generalised anxiety, social anxiety, specific phobias and panic disorder, and conduct disorder has its own recognised subtypes. They also note that only some people with a given disorder may show brain differences, so averaging across the whole group may yield small effects. That is a reason to be curious about subgroups. It is not a basis for a scan-based test.

The practical point for families is plain. No MRI, at these effect sizes, can say whether a child has ADHD, and the paper does not suggest one could. A diagnosis is still made by clinical assessment, the route set out in the Directory’s guide to getting an ADHD diagnosis in the UK, and the providers who carry it out are listed under assessments. The same caution applies to small studies reporting large effects. The Directory’s piece on a structural MRI study of brain asymmetry in 40 children and adolescents with ADHD and 30 controls reported its own, larger effect sizes on a different measure. The two are not directly comparable, but a sample of 70 and a sample of 8,780 do not carry the same weight.

Shared axes, not separate brains, and where that reading stops

The study’s design starts from a familiar model that sorts youth disorders into internalising and externalising groups, and its results press against even that. No pattern belonged to the internalising disorders alone. The most consistent findings cut across all four diagnoses. And none of the disorder-specific findings met the authors’ test for exclusivity.

That fits a frame the Directory has argued from the other direction. Chapter V of The Neurodiversity Book, “Universal Cognitive Mechanisms”, reads ADHD and autism criteria as positions on shared axes rather than separate kinds of person, and puts the conclusion this way:

The diagnosis names that crossing. It does not name a different kind of brain, because there is no different kind of brain to name, only a different tuning of the same one, and a line someone drew across it.

The Neurodiversity Book, second edition, chapter V

The fit is partial. The book’s argument is about cognitive mechanisms, and its pairing is ADHD and autism. This paper measures brain anatomy, does not include autism, and sets ADHD beside conduct disorder, anxiety and depression. It does not test the book’s axes and should not be read as confirming them. Nor does it find the groups identical. The authors report differences in degree between groups, and in some regions opposite effects: conduct disorder and anxiety differed from controls in opposite directions in parahippocampal surface area, and differed significantly from each other. What it shows is narrower and, for the categorical model, harder to dismiss. Measured together, at scale, four diagnoses that are treated as distinct kinds produced mostly shared, mostly small and nowhere exclusive differences. That is consistent with diagnoses as thresholds drawn across common variation. It is a poor fit for diagnoses as windows onto distinct brains.

It sits beside work the Directory has covered on function rather than anatomy. One review examined executive function and emotion regulation as proposed transdiagnostic mechanisms across ADHD and autism. A Padova study found that functional profiles cut across the autism and ADHD diagnostic boundary. A separate mega-analysis of cortical thickness and curvature reported opposite developmental trajectories for autism and ADHD; with different measures, ages and atlas, it answers a different question.

The structural problem is that diagnosis does work this evidence cannot support. It gates support and it fixes a child’s file. Where it is explained as a difference of brain, the evidence of what its authors call the largest cross-disorder analysis of brain structure in youths to date says that explanation is ahead of the anatomy: the differences are small, mostly shared, and in ADHD partly entangled with the prescription that follows the diagnosis. The DWP’s Expert Academic Panel on Neurodiversity has already argued that support at work should follow need, not diagnosis. If the brain does not draw the line between these categories, the case for letting the category decide who gets help is weaker still. The test now is whether the services that rely on that line can show what it measures that need does not.

Citations

Townend, S., Staginnus, M., Gao, Y. et al. (2026) — Shared and Distinct Alterations in Brain Structure of Youth With Internalizing or Externalizing Disorders: Findings From the ENIGMA Antisocial Behavior, ADHD, Major Depressive Disorder, and Anxiety Working Groups — Biological Psychiatry

News and updates.

Straight to your inbox.