4,996 adults, one dataset, and a pattern the clinical apparatus refuses to read
Luca Hargitai, Lucy Waldren, Lucy Livingston, Florence Leung and Punit Shah at the University of Bath, King’s College London and the University of the West of England published an open-access paper in Scientific Reports in April 2026 examining the unique contributions of autism and ADHD to internalising problems — generalised anxiety disorder and depression — in adulthood. The dataset is substantial. A general population sample of 4,996 adults drawn from the United Kingdom and the United States. A matched clinical sub-sample of 292 participants — 73 with a clinical autism diagnosis only, 73 with a clinical ADHD diagnosis only, 146 neurotypical controls — matched on age, sex and education level. A further screening-based sample of 1,728 participants — 432 probable autism, 432 probable ADHD, 864 neurotypical controls — also matched. Three different ways of looking at the same architectural question.
The trait-level finding from the general population sample is straightforward. Both autistic traits and ADHD traits independently predict internalising problem diagnoses. ADHD traits are the stronger predictor — each unit increase in ADHD traits raises the odds of an internalising diagnosis by 4%, against 2% per unit increase in autistic traits. Both effects are highly statistically significant at this sample size. The general population pattern is unambiguous: more neurodivergent traits, more anxiety and depression.
The clinical sub-sample analysis is where the pattern starts to speak architecturally. Both autistic adults and adults with ADHD have substantially elevated odds of internalising problems compared to neurotypical controls — odds ratio 3.94 for the autism-only group, 5.00 for the ADHD-only group. The two clinical groups don’t statistically differ from one another in their overall likelihood of having an internalising diagnosis. But the type of internalising diagnosis is patterned. Adults with an ADHD diagnosis are more closely linked with depression (OR 3.81) than autistic adults are (OR 3.41). Autistic adults are more closely linked with generalised anxiety disorder (OR 4.64) than adults with ADHD are (OR 3.35). The differences between the clinical groups aren’t statistically significant — but the pattern of asymmetry is the architectural signal.
The probable-group analyses, using validated screening thresholds rather than clinical diagnoses, sharpen the picture further. Adults meeting screening criteria for probable ADHD are significantly more likely to have internalising diagnoses than both the neurotypical control group AND the probable autism group. Adults meeting screening criteria for probable autism, after correction for multiple comparisons, are not significantly more likely to have internalising diagnoses than neurotypical controls. The clinical autism group elevation appears to be driven by something the screening doesn’t capture — most likely the substrate-environment friction that produces clinical-level distress in autistic adults whose substrate the apparatus has already recognised, contrasted with the screening-positive autistic adults whose substrate is still operating without that recognition.
Two further numbers worth holding. Michael Hollocks and colleagues’ 2019 meta-analysis, which Hargitai et al. cite as their baseline, estimates depression and anxiety prevalence in autistic adults at 37% and 42% respectively. The general population prevalence in the UK and US sits around 8% for depression and 7% for generalised anxiety disorder. The neurodivergent multipliers are not small. They are not within sampling error. They are the structural signature of something happening at population scale that the clinical apparatus has been describing in the wrong language for decades.
"Shared neurobiological vulnerability" is the wrong story
The standard clinical framing the Hargitai paper opens with — and treats as the default account it is responding to — is that autism, ADHD and affective disorders share intrinsic neurobiological pathology. The co-occurrence is read as evidence of shared underlying disease. Different conditions, same biology, overlapping vulnerability. This is the reading mainstream psychiatry has been operating with, with variations, for thirty years. It justifies the diagnostic category architecture, the medication pathway, the treatment manualisation, and the research agenda that goes looking for shared genetic and neural substrates.
The reading is wrong, and the Hargitai data is one of several recent datasets making it harder to maintain. The pattern of differences the paper documents — ADHD differentially predicting depression, autism differentially predicting anxiety, the two substrates producing different internalising signatures — is not what shared underlying pathology predicts. Shared underlying pathology predicts shared phenotypic expression. What you would expect to see, if the clinical reading were correct, is roughly similar rates of anxiety and depression across both neurodivergent groups, driven by the same upstream vulnerability. What you actually see is two substrates producing two distinct affective patterns, both elevated against the neurotypical baseline, but with internal asymmetry that maps onto how each substrate meets the environment.
The paper’s own mechanistic speculation tells the architectural story without quite naming it. The ADHD-internalising link the authors trace to “response inhibition difficulties” at the trait level and “emotional dysregulation and irritability” at the clinical level. The autism-anxiety link they trace to “intolerance of uncertainty” and “strong preference for predictability.” Each of these is presented as if it sits inside the substrate as a deficit-feature producing downstream affective consequence. Re-read them structurally. Response inhibition difficulties manifest as problems with sustained compliance against externally imposed tasks the substrate is not designed to maintain at the required cadence. Emotional dysregulation manifests as affect that exceeds the dynamic range the demand structure permits in workplaces, schools, and healthcare settings. Intolerance of uncertainty manifests as distress in environments that the substrate cannot anticipate, in settings whose calibration assumes the autistic substrate can process unpredictability the way a different substrate would. These are not deficit-features. They are substrate-features colliding with a demand structure calibrated against them. The “mechanism” the paper describes is the friction itself, named individually rather than structurally.
The paper acknowledges the structural reading exists. Reference 75 in their citation list points to the social model of disability. Their discussion mentions “important environmental modifications that could be made to improve wellbeing in autism and ADHD” and refers explicitly to “societal failure to accommodate the physical, social and emotional needs of autistic people.” But this acknowledgement sits as a parenthetical alongside the individual-mechanism account, treated as one consideration among several rather than as the load-bearing explanation the data actually demands. The structural account is mentioned. It is not centred. The Directory’s position, here as elsewhere, is that the structural account is not optional and not parenthetical. It is the only reading the differential pattern of internalising outcomes actually supports. Shared neurobiological vulnerability would produce shared phenotypic expression. Shared friction with a demand structure that rejects each substrate differently produces differentiated phenotypic expression. The data shows the second.
What this matters for, beyond the academic argument: the intervention pathway downstream of the diagnostic apparatus follows the reading. If anxiety and depression in neurodivergent adults are read as shared underlying pathology, the intervention targets the affective surface — CBT, SSRIs, anxiolytics, and behavioural treatments. The substrate continues to meet the demand structure. The friction continues to accumulate. The cost continues to be paid. If anxiety and depression are read as the cost of substrate-environment friction, the intervention targets the structure rather than the substrate. The substrate is recognised, the demand structure is redesigned to meet it, and the affective cost stops being chronically generated. The two readings produce fundamentally different healthcare systems. The current one is calibrated for the first reading. The data, increasingly, supports the second.
What lifelong friction with the round hole actually does to a developing self
The ADHD and autism substrates operate in the world the cybernetic demand structure has constructed differently from one another, and each finds itself repeatedly producing outputs the structure registers as failures — in two distinct patterns. Start with ADHD. Tasks that require sustained attention against externally imposed cadence — meetings, lectures, deadlines, sustained reading, sustained writing, sustained anything against a clock not the substrate’s own — generate non-compliance. The substrate produces non-compliance because the substrate isn’t built for sustained externally-paced attention; it samples broadly, attends to salience as salience changes, integrates affect with cognition rather than separating them, and operates on intrinsic rather than externally-imposed rhythm. The structure interprets the non-compliance as the substrate’s failure to perform what the structure considers basic. Repeated across a developmental arc — primary school, secondary school, university, workplace — the substrate accumulates thousands of small “failure” events. Forgotten appointments. Missed deadlines. Unfinished tasks. Conversations attended-to inadequately. Promises made and not kept. Each event is registered, by the surrounding adults and by the substrate itself, as evidence of personal inadequacy.
This accumulation does specific architectural work. Each failure event produces a small dose of shame. The shame compounds. Self-concept, then, develops around (and has a tendency to seek reinforcement for) this accumulating evidence of “failure.” The internal narrative ultimately consolidates: I am lazy, I am unreliable, I am letting people down, and something is wrong with me (or something along those lines). The cognitive style of this accentuated default mode network — recursively biographical, self-referential, prone to rumination and mind wandering — folds the accumulating shame back on itself, ad infinitum. The substrate produces, eventually, learned helplessness about whether sustained externally-paced compliance is achievable at all. Effort starts to feel pointless because effort has historically failed to produce reliable compliance. The affective signature of this accumulation is depression. Not depression as a separate disease running on its own substrate. Depression as the predictable downstream consequence of a substrate that produces what it produces, meeting a demand structure that has calibrated what counts as adequate performance against a different substrate, across decades of repeated failure-to-comply regardless of effort and willpower. The Hargitai data showing ADHD adults specifically elevated for depression is this pattern made statistical.
The autism substrate operates in the same demand structure differently and produces a different friction signature. The substrate needs predictability — needs the environment to behave consistently enough that the substrate can model it, prepare for it, and configure regulatory capacity against it in advance. Schools, workplaces, social settings, healthcare environments, parenting, family events, social plans — most of contemporary life is structurally unpredictable in ways the substrate cannot model. Plans change without notice. Sensory environments shift without warning. Social rules update without being named explicitly. Tone changes mid-conversation. The substrate operates in chronic uncertainty about whether the next minute will be tolerable, whether the next interaction will land safely, and whether the next environment will impose sensory inputs the substrate could struggle to process. This chronic uncertainty produces chronic threat-detection. The threat-detection system is not over-active or hypervigilant in some context-free sense. It is — again, learned behaviour inside the nervous system — calibrated correctly for an environment that does not, in fact, behave predictably for this substrate. The affective signature of running a correctly-calibrated threat-detection system against a chronically uncertain environment is anxiety. Not anxiety as a separate disease. Anxiety as the predictable downstream consequence of a substrate that requires coherent predictability meeting a demand structure that does not provide it. The Hargitai data showing autistic adults specifically elevated for generalised anxiety disorder is this pattern made statistical.
Two substrates (that, when inventoried, are a near-comprehensive list of active universal cognitive mechanisms; “ADHD” and “autism” being the clinical diagnostic apparatus that catches threshold-firing beyond the flat-lined expectation of typical). One demand structure. Two friction signatures. One downstream affective cost (“internalising symptoms”). The clinical reading splits the cost into separate “disorders” and treats them as such, which is structurally what the diagnostic apparatus is designed to do — fragment the response, name each fragment, treat each fragment, and leave the producing structure unexamined. The Hargitai data, read structurally, refuses the fragmentation. The differential pattern of internalising outcomes is the architectural signature of substrate-environment mismatch, distributed differently across two substrates, producing two affective signatures, neither of which is intrinsic to the substrate or attributable to a shared underlying disease. The “comorbidity” the apparatus treats is the cost of the friction the clinical apparatus cannot do anything about.
I lived this directly. I was misdiagnosed with severe anxiety and depression at fifteen — six years before the substrate underneath was finally recognised. The GP who diagnosed me wasn’t doing anything wrong by the standards of GP practice at the time. Any other fifteen-year-old that scored how I scored (over the threshold) on the paper I had to tick boxes on get diagnosed with depression and anxiety. The substrate underneath isn’t part of the diagnostic conversation, because the apparatus the GP is operating inside doesn’t include “the cognitive substrate is a poor fit for the environment it’s been operating in for fifteen years” as a recognised diagnostic possibility. So the symptoms got treated. The substrate continued to meet the same demand structure. The accumulation continued. The eventual recognition of AuDHD at twenty-one didn’t undo the previous six years of treatment-without-substrate-recognition; it simply named what had been happening underneath the symptoms the whole time, and allowed me to get off the SSRI merry-go-round and jump on a different pharmacological ride (a seesaw!).
My partner, Lucy, is also neurodivergent: currently undiagnosed and untreated. I observe what the lifelong friction does to a self that hasn’t yet been given the framework to read its own history. The pattern is the same architecture. A substrate that produces what substrates produce, meeting environments that have never been calibrated for it, across decades of accumulated misreading by every system the substrate has had to interact with — school, family, healthcare, workplace, social settings, etc. The internalising — anxiety and depression and how they show up in the internal experience of the self — is the cost being paid in real time. Lucy isn’t experiencing pathology. Lucy is experiencing a substrate that hasn’t been recognised meeting a demand structure that doesn’t recognise it, with the predictable affective signature accumulating where the recognition should have been. The case of my partner and I are but two cases.
Neither of these cases is unusual. The Hargitai dataset is the population-level signature of cases like these — at scale, across the United Kingdom and the United States, in the hundreds of thousands of neurodivergent adults whose anxiety and depression have been treated as primary conditions when what they actually are is the cost of lifelong substrate-environment friction. The apparatus catches each individual case as if it were singular, treats the affective surface, and discharges the patient with the substrate still meeting the structure that produced the cost. The cost continues to accumulate. The next presentation is recorded as either treatment-resistant depression or recurrent generalised anxiety disorder. The apparatus has no diagnostic category for “what’s actually happening here is that the substrate hasn’t been recognised and the demand structure hasn’t been redesigned around it.” So it doesn’t catch it. So it can’t intervene against it. So the cost keeps being paid.
What this means for late-diagnosis adults and the apparatus that catches them
Late-diagnosis adults — the rapidly growing cohort whose ADHD or autism or AuDHD was identified in adulthood after years or decades of operating without recognition — already know what the Hargitai data is now showing at population scale. The anxiety and depression that brought us to clinical attention in our teens or twenties was not pathology of the substrate. It was the cost the substrate was paying for meeting the environment without recognition. The cost was real. The clinical conditions were real. The substrate underneath was also real, and the apparatus didn’t see it. The apparatus saw the cost, recorded the cost as the condition, treated the cost as the condition, and discharged the patient with the substrate still operating without recognition. The substrate kept producing the cost. The condition kept recurring.
This pattern is endemic in late-diagnosis trajectories. Read any large sample of late-diagnosis adults’ clinical histories and the structure is repetitive. Initial mental health presentation in teens or early twenties — anxiety, depression, eating difficulties, OCD, sometimes a personality disorder label. Treatment courses delivered against these labels. Partial response, relapse, treatment escalation, partial response, relapse. Sometimes one or more “treatment-resistant” specifications added to the diagnosis. Sometimes additional comorbid diagnoses added — bipolar, complex PTSD, borderline personality disorder — as the apparatus tries to account for why the original conditions don’t resolve under standard treatment. The substrate, all the while, is doing what substrates do, meeting environments calibrated against it, and accumulating the cost. The eventual recognition of the underlying neurotype, when it comes, recontextualises the entire history. The conditions that were treated were not the substrate. They were the cost the substrate was paying. The substrate was always doing what it did. The apparatus was looking in the wrong place.
What this means for clinical practice is not subtle and not negotiable. Adults presenting with anxiety and depression — especially treatment-resistant, recurrent, or with a presentation pattern that doesn’t fit standard trajectories — need to be assessed for the underlying neurodivergent substrate before the affective conditions are treated as primary; so that the individual can re-gain and be able to hold a coherent sense of self without the baggage of pathology. The current standard pathway treats the substrate assessment as an afterthought, a specialist referral, a thing considered only after standard treatments have visibly failed. The data Hargitai et al. produce, and the broader empirical pattern the data sits inside, requires the substrate question to be asked first. If the substrate is neurodivergent, the anxiety and depression are likely the cost of substrate-environment friction, and the intervention pathway has to address the structural mismatch, not just the affective surface. If the substrate isn’t neurodivergent, the standard treatments may be appropriate. Either way, the question has to come first. Currently it comes last, if at all.
What this means for late-diagnosis adults still inside the apparatus is also not subtle. The framework that explains your clinical history is structural. The anxiety and depression you were diagnosed with at fourteen, sixteen, nineteen, twenty-five, whatever, was not the substrate (your nervous system; your neurology). It was the cost your substrate was paying, and you never knew why. The substrate itself — the human being — is not the problem. The friction is. The recognition of the substrate is the beginning of the intervention for how to best deal with that friction; it is not the intervention itself. The intervention is the demand structure changing to be better suited, the environment being calibrated for the substrate that meets it, and, most importantly, the accumulated friction being allowed to stop accumulating. This is not what the standard pathway offers. The standard pathway offers continued symptom management for the cost while the friction continues. The structural alternative is what my corpus has been arguing across every domain — and what the Hargitai data, taken seriously and read structurally, vindicates as the only reading the empirical evidence actually supports.
The paper’s authors stop short of saying this. Their discussion gestures at the social model in a parenthetical clause, mentions environmental modification as a future research consideration, and concludes with a call for more research into psychological mechanisms underpinning the relationships. They are operating inside the constraints of a peer-reviewed psychiatric journal whose editorial expectations require the individual-mechanism account to remain the centre of gravity. My corpus, thankfully, operates outside those constraints. I can, then, read what the data shows. The data shows that two substrates, meeting one demand structure that rejects each of them differently, produce two distinct affective signatures whose pattern is the structural fingerprint of the friction itself. The “shared neurobiological vulnerability” reading was always the wrong story. The data has been telling us for years. The Hargitai paper is the latest dataset to tell us again. The clinical apparatus has not yet listened, and it probably won’t. The cost continues to be paid in the lives of every neurodivergent adult whose substrate the apparatus hasn’t yet recognised.
Citations
Hargitai, L. D., Waldren, L. H., Livingston, L. A., Leung, F. Y. N. & Shah, P. (2026) — Neurodiversity and mental health in adulthood: exploring the unique contributions of autism and ADHD to internalising problems — Scientific Reports 16:16343
Hollocks, M. J., Lerh, J. W., Magiati, I., Meiser-Stedman, R. & Brugha, T. S. (2019) — Anxiety and depression in adults with autism spectrum disorder: a systematic review and meta-analysis — Psychological Medicine 49:559–572
Hargitai, L. D., Livingston, L. A., Waldren, L. H., Robinson, R., Shah, P. (2023) — Attention-deficit hyperactivity disorder traits are a more important predictor of internalising problems than autistic traits — Scientific Reports 13:31
Lai, M. C., Kassee, C., Besney, R., Bonato, S., Hull, L., Mandy, W., Szatmari, P. & Ameis, S. H. (2019) — Prevalence of co-occurring mental health diagnoses in the autism population: a systematic review and meta-analysis — The Lancet Psychiatry 6:819–829
Sonuga-Barke, E. & Thapar, A. (2021) — The neurodiversity concept: is it helpful for clinicians and scientists? — The Lancet Psychiatry 8:559–561
Zarb, G. (1995) — Modelling the social model of disability — Critical Public Health 6:21–29
