Overshadowed: Does ADHD Explain Late-Diagnosed Autism?

The question

Late diagnosis of autism in adulthood is rising, and AuDHD - co-occurring autism and ADHD - is common enough that pooled estimates put it around two in five autistic people, with individual studies ranging far wider than that pooled figure suggests.¹ Is a co-occurring ADHD diagnosis part of why the autism diagnosis arrived late - and if so, how much of the trend does it actually explain, as opposed to merely accompany?

As far as I can find, nobody has measured this directly in adults.²

Where the question came from

I wrote in my book Loving Variant Minds about the rising trend of late autism diagnosis and the high propensity for comorbid ADHD, and moved on without asking the obvious next question: why is it these particular people, so often, and not others? Part of the standard answer is well rehearsed - pre-DSM-5 diagnostic categories that missed anyone without visible impairment, camouflaging, clinicians trained on a young-boy-shaped prototype. All of that is true and none of it is new.

What is less rehearsed is the second condition sitting right next to the first one. ADHD and autism co-occur constantly. A meta-analysis of sixty-three studies put the pooled current prevalence of ADHD in autistic people at 38.5% and the lifetime prevalence at 40.2%, and noted that the individual studies feeding into that average varied by age, IQ, recruitment setting, and which diagnostic criteria were used - which is a polite way of saying the true figure depends heavily on where you go looking.¹ Popular accounts routinely quote a much wider span than that pooled figure, and while I cannot trace a precise upper and lower bound to a single peer-reviewed source, the direction is not in dispute: this is a large, heterogeneous, commonly co-occurring population, not an edge case.

So the question is not exotic. Two things are true - a rapidly growing number of autism diagnoses are happening in adulthood, and a large fraction of autistic adults are also ADHD - and it would be strange not to ask whether the second fact is doing some of the work of the first.

Why the prevalence figure alone proves nothing

The obvious move is to point at that 38-to-40% figure and treat it as the answer: a lot of autistic people have ADHD, therefore ADHD is clearly a major driver of late diagnosis. I want to head this off before the rest of the proposal, because it is the same mistake as reaching for a divorce rate to measure relationship damage - a real number, aimed at the wrong question.

A prevalence rate tells you how often two things co-occur. It tells you nothing about timing, and timing is the entire question here. If ADHD-plus-autism adults were diagnosed with autism at exactly the same age as autism-only adults, the co-occurrence could be as high as you like and would explain none of the lateness. The number that would actually answer the question is a comparison: age at autism diagnosis in the co-occurring group versus the autism-only group, holding everything else equal. Nobody has quoted that comparison to me, because as far as I have found, nobody has produced it for adults.

There is a second problem with the raw prevalence number, and it is the one the meta-analysis itself flags: recruitment setting moves the estimate.¹ A study drawing from an ADHD clinic will find autism at whatever rate that clinic happens to screen for it, and a study drawing from an autism clinic will do the reverse. A high co-occurrence rate can therefore be partly an artefact of who gets asked the second question, which is a fact about clinical practice rather than a fact about the two conditions - and it is exactly the kind of contamination that a study of why diagnosis is late has to design its way around rather than inherit.

What the evidence actually says

The children’s literature is real, replicated, and measured in the wrong units

Here the finding is solid. Kentrou and colleagues found that children and adolescents with a prior ADHD diagnosis had their autism recognised, on average, 1.5 years later than those without one - and 2.6 years later for girls specifically.³ Sainsbury and colleagues found something structurally similar from the other direction: children who went on to receive both diagnoses got their ADHD diagnosis earlier than ADHD-only children (a median of 79 months against 96.2) and their autism diagnosis substantially later (119.6 months against 81), with parents of the co-occurring group reporting less obviously atypical language and social development in the years before diagnosis.⁴ A 2026 analysis of over 13,000 children on Missouri Medicaid found that a larger cumulative count of prior diagnoses predicted an older age at autism diagnosis; ADHD was the single most common of those prior diagnoses, though the analysis did not show it carrying more of that effect than any other diagnosis category.⁵ Most recently, a 2025 study found that three or more co-occurring psychiatric conditions delayed autism diagnosis by four to six years, with ADHD and depression among the specific delaying factors.⁶

Four independent findings, one direction, real effect sizes. I have no quarrel with this literature.

My quarrel is with what it is a study of. Every one of these delays is measured in months or a handful of years, within childhood - Kentrou’s 1.5-to-2.6-year gap, Sainsbury’s roughly three-year gap in the co-occurring group. That is not the same phenomenon as the one this proposal is actually about: a diagnosis arriving not a few years late but three or four decades late, to someone who was never assessed at all until adulthood. Nobody has shown that the mechanism which nudges a childhood diagnosis from six to nine is the same mechanism, run for longer, that keeps a diagnosis from arriving until forty-five. It might be. It is also entirely possible that the pediatric delay and the adult delay are different phenomena that happen to share a name, in much the way autism and alexithymia turned out to share a symptom for years before anyone pulled them apart, as I found when I looked into interoception literature. Extrapolating a two-year childhood effect across four decades of adulthood is the size of assumption a dissertation is supposed to test, not make.

The genetic and neural evidence is stronger than a literature search would lead one to expect

I did not expect to find this, and it is the best evidence for the whole proposal - not one finding but three, converging from different methods.

Zhang and colleagues analysed genetic and developmental data across four birth cohorts and found two distinguishable trajectories associated with age at autism diagnosis.⁷ One trajectory is linked to earlier diagnosis, lower early-childhood social and communication ability, and only a moderate genetic correlation with ADHD. The other is linked to later diagnosis, socioemotional and behavioural difficulties that emerge more in adolescence than early childhood, and a moderate-to-high genetic correlation with ADHD and other mental health conditions. That is a genuinely different kind of evidence from the clinical-delay literature above, because it is not about who got referred to whom. It suggests that late-diagnosed autism and ADHD-linked autism may not just co-occur by chance encounter in the clinic; they may share developmental and genetic architecture.

Two more recent studies push in the same direction using a completely different method, and between them they sharpen the phenotype argument rather than simply restating it. Watanabe and Watanabe analysed resting-state brain dynamics in 338 children and found that ASD+ADHD is not a simple merger of the two conditions: the social-communicative signature in the comorbid group ran on exactly the same neural mechanism as in autism alone, while the ADHD-like traits ran on a different mechanism entirely - atypically frequent brain-state transitions driven by the frontoparietal control network and left prefrontal cortex, rather than the dorsal-attention-network and parietal instability that characterises ADHD on its own.⁸ A far larger mega-analysis by Norman and colleagues, pooling resting-state data from 12,732 children and adolescents, reached a compatible conclusion at scale: autism and ADHD traits are associated with distinct patterns of functional connectivity even where they co-occur, though the effect sizes involved are small.⁹

Read those two findings side by side and they cut against the simplest version of the phenotype account rather than confirming it outright. If autism in AuDHD were merely fainter or harder to detect, its neural signature would look attenuated in the comorbid group. It does not - Watanabe and colleagues found it identical to pure autism. What differs is the ADHD half of the presentation, which behaves enough like ADHD to be read as ADHD while running on different machinery underneath. That is closer to a biological account of why overshadowing would be easy to fall into than it is evidence that the autism itself presents differently: on this evidence, the marker most relevant to an autism diagnosis is fully intact and simply not being asked about once a plausible-looking ADHD has already closed the file.

Both of the newer samples are children again - 338 in the first study, six-to-nineteen-year-olds in the second - so the same objection about units applies here as everywhere else in this section. But it is now two independent methods, genetic and neuroimaging, agreeing that AuDHD is not additive. That is a sturdier foundation than either alone, and it changes what the “distinct phenotype” branch of this proposal is actually claiming: not that autism looks weaker in AuDHD, but that ADHD looks different enough in AuDHD to plausibly absorb the clinician’s whole attention while a fully-formed autism sits underneath it, unasked-about. If that holds up, “is AuDHD a contributor to late diagnosis” stops being only a question about clinician behaviour and becomes a question about which of at least two biologically distinguishable presentations of autism a given late-diagnosed adult actually has - and, per the neural evidence, why the second one is so easy to talk yourself out of looking for.

The adult literature has the prevalence but not the timing

Turn to adults and the picture inverts: plenty of data on how often the two conditions co-occur, almost nothing on when. A large Medicaid study found ADHD rates in autistic adults sharply elevated relative to the general population, with co-occurrence linked to worse health outcomes.¹⁰ A study of newly diagnosed Greek adults found at least one co-occurring psychiatric condition in half of the ASD-only group and nearly three-quarters of the ADHD and ADHD-plus-ASD groups,¹¹ with a companion analysis finding that the combined group is marked out from the ADHD-only group by reduced imagination and empathy alongside heightened attention to detail, and marked out from the ASD-only group by a different pair of traits entirely - childhood hyperactivity and current impulsivity.¹² The prospective ABIS study of young Swedish adults found that co-occurring ADHD and autism carried a heavier combined burden than either alone: higher unemployment, a tendency toward autoimmune conditions, and a psychiatric profile that resembled ADHD while a behavioural profile - notably lower risk-taking - resembled autism.¹³

Not one of these studies asks when the autism diagnosis arrived relative to the ADHD diagnosis, or whether that gap differs from the ASD-only group. They establish that AuDHD adults exist in large numbers and fare worse. They do not establish that AuDHD adults wait longer for the autism half of that pair. And none of them has looked for the adult analogue of what Watanabe and Norman found in children - whether the same neural dissociation holds once the population has actually reached the age this proposal is about.

The mirror-image phenomenon, and why it doesn’t answer this question

Community discourse about AuDHD often states the relationship symmetrically: ADHD hides autism, and autism hides ADHD. The first half is what this proposal is about, and there is a second, larger piece of evidence for it beyond Kentrou above. Miodovnik and colleagues used a population-based sample of nearly 100,000 US children, including 1,496 with autism, and found that 45% of those who also had ADHD were diagnosed with ADHD first - and that group waited roughly three years longer for the autism diagnosis (95% CI 2.3–3.5) and were close to thirty times more likely to be diagnosed with autism after age six.¹⁴ That is a second, larger, independently sourced confirmation of the same direction as Kentrou et al., built on parent report rather than clinical record, a limitation the authors themselves flag against reading it as causal.

The second half of the community claim - autism hides ADHD - is also true, but for a different reason than symptom overlap, and it answers a different question than the one this proposal asks. From 1994 to 2013, the DSM-IV explicitly excluded ADHD as a diagnosis in anyone who already had a pervasive developmental disorder diagnosis - not a clinician’s judgement call, a rule written into the manual.¹⁵ DSM-5 removed the exclusion in 2013, in recognition of how often the two conditions actually co-occur. That means an entire generation of people diagnosed autistic as children before 2013 - a substantial share of today’s late-diagnosed autistic adults, as it happens - could not have had ADHD written down at the time no matter how clearly it presented, and many will never have been reassessed since. A live clinical version of the same asymmetry persists after the rule changed: Rau and colleagues found that standard rating scales are specifically weak at telling the inattentive presentation of ADHD apart from autism-related attention difficulties, markedly weaker than for the hyperactive-impulsive presentation - so even under DSM-5’s permissive rules, the quieter form of ADHD is still the one most likely to go unnoticed inside an existing autism diagnosis.¹⁶

Both halves of the community claim are backed by real evidence, in other words, but they are not the same study waiting to be run. This proposal is about why autism arrives late. The DSM-IV history and the Rau finding explain why ADHD arrived late, or never arrived, in people already diagnosed autistic - the mirror image of this proposal’s question, resting on a mechanism more decisive than anything else in this proposal, since a diagnostic exclusion rule beats a clinical hunch every time - but it is not evidence for this proposal’s actual claim. Anyone running the vignette or registry arms below should keep the two questions separate, because a result that really reflects the DSM-IV history would be mistaken for evidence of overshadowing if the two got pooled.

The part where I argue against myself

I would rather state the weaknesses in my own idea than have a supervisor state them for me. There are four.

The units problem, again. I raised it above and it is serious enough to repeat as an objection rather than a caveat: nothing in the pediatric, genetic, or neuroimaging literature has been tested against the population this proposal actually cares about - adults diagnosed in their thirties, forties, and fifties. A study that simply assumes the childhood mechanism scales up is assuming its own conclusion.

Overshadowing, masking, and a genuinely different phenotype are three different claims wearing one name. “AuDHD explains late diagnosis” could mean any of three things, and they have different causes and different fixes. It could mean diagnostic overshadowing: a clinician has an ADHD diagnosis on the chart, attributes the remaining traits to it, and stops looking - a fact about clinician behaviour, fixable by clinician behaviour. It could mean camouflaging: something about having both conditions - higher verbal ability, more compensatory strategy, more practice performing normally - makes the autism harder for anyone to detect, which is close to what Milner and colleagues found when they showed that camouflaging predicts age at autism diagnosis, with a stronger relationship for women than men.¹⁷ Or it could mean a genuinely different developmental phenotype - the Zhang et al. trajectory in which the difficulty simply does not present the same way in early childhood, so there is nothing early to overshadow or camouflage in the first place.⁷ A study that finds “AuDHD adults are diagnosed later” and stops there has found a correlation compatible with all three explanations and has recommended nothing, because the three point at different fixes: clinician training, better instruments, or a revised diagnostic category.

The neural evidence above sharpens the overshadowing account rather than sitting outside it. Watanabe and colleagues found that the autism-specific signature in AuDHD children was neurally identical to pure autism - not fainter, not attenuated - while the ADHD-like presentation ran on different machinery that nonetheless looks enough like ADHD to be diagnosed as ADHD, correctly.⁸⁹ That matters, because it means overshadowing does not require a careless clinician. A textbook-looking ADHD presentation can be read entirely correctly and still leave a fully-formed, fully-detectable autism sitting underneath it, unasked-about, because diagnosing the first condition well does not require querying the second. Call it diagnostic sufficiency rather than diagnostic error: a more forgiving version of overshadowing than “the clinician missed it,” and a more testable one, because it predicts that Arm One below should find an effect even among careful, well-trained clinicians - the omission would be structural, not a lapse in judgement.

“Late” does not mean one thing. A 2025 systematic review of the late-diagnosis literature found that authors’ cutoffs for what counts as “late” ranged from age two to age fifty-five, with a bimodal split clustering around three and around eighteen, and concluded that the field has no working consensus on the term.¹⁸ This is not a side issue. If the pediatric studies above call a diagnosis “late” at age nine and the adult literature calls a diagnosis “late” at age forty-five, a design that borrows the word without checking what it is standing for will produce a number that looks precise and means two different things depending on which paper you read next to it.

Gender complicates the mechanism rather than confirming it. Gesi and colleagues found that misdiagnosed autistic men were most often labelled with ADHD, while misdiagnosed autistic women were more often labelled with a personality disorder - though the misdiagnosed subgroups behind that finding were small (seven men, ten women), and ADHD comorbidity in their sample overall was low, at under five per cent.¹⁹ If overshadowing by ADHD specifically is a male-coded pathway and a different overshadowing entirely - by personality disorder - delays women, then “does AuDHD explain late diagnosis” needs a gender-stratified answer from the outset, not a pooled one that averages across two different delaying mechanisms.

There is a further layer here, and it is worth stating cautiously rather than confidently. Girls and women with ADHD are consistently overrepresented in the inattentive presentation rather than the combined or hyperactive-impulsive one, and a review by Hinshaw and Nguyen concludes that this quieter presentation is itself part of why ADHD in girls goes underidentified.²⁰ That connects to Rau and colleagues above: the inattentive presentation is specifically the one current instruments are worst at telling apart from autism.¹⁶ Set that against what is already established in this section - camouflaging predicting age at diagnosis more strongly in women,¹⁷ and longer diagnostic delays for women generally¹⁹ - and there is a plausible compounding pathway: women with AuDHD may be more likely to carry the ADHD presentation that is hardest to separate from autism, on top of an already-documented camouflaging effect.

I want to be precise about what that is and is not evidence of. It is not evidence that women’s inattention is inherently quieter or harder to detect at the level of biology. The more rigorous test of that question - clinical diagnostic interview rather than rating scales - found no significant gender difference in inattention or hyperactivity severity at all; the gender gap appears only in rating-scale and community data, which may reflect referral bias as much as anything true of the person.²¹ What differs by gender, on the best evidence, is not the underlying symptom but which presentation category it gets sorted into and how visible that category is to the people doing the sorting - and it is that sorting, not an essential difference between men and women, that this proposal would need to test for. The same caution applies to the autism gender ratio itself: the historical assumption of four boys diagnosed for every girl has not held up. The most careful meta-analysis puts the true ratio closer to three to one, driven substantially by which studies actively look for girls rather than only counting those already diagnosed,²² and the most recent birth-cohort data show that ratio continuing to narrow, approaching parity in the youngest cohorts studied.²³ Any gender-stratified analysis in this proposal should expect a gap that is real but narrower than assumed, and should be built to find out why it exists rather than to assume what it is.

The design that solves the hard problem

The observational data cannot separate overshadowing, camouflaging, and genuine phenotype, because in any real patient all three are entangled with each other and with the very fact that they are already diagnosed. The way out is to stop trying to disentangle them after the fact and instead hold the patient constant while manipulating only one variable at a time.

Arm one: a clinician vignette experiment, to isolate overshadowing. Present qualified clinicians with identical case descriptions of an adult presentation - same traits, same history, same complaint - and randomise only whether the case notes mention a pre-existing ADHD diagnosis. If clinicians presented with the ADHD-labelled version are less likely to pursue an autism assessment, refer more slowly, or hold a higher threshold before referring, that isolates overshadowing specifically, because the only thing that changed was the label, not the person.

Arm two: a camouflaging and cognitive battery, to isolate masking. Recruit adults with and without a pre-existing ADHD diagnosis at the point of autism referral, administer the Camouflaging Autistic Traits Questionnaire alongside a cognitive-ability measure, and test whether camouflaging score - not diagnostic label - predicts age at referral within the ADHD-positive group. If camouflaging is doing the work, it should mediate the relationship between ADHD status and referral age; if it does not, masking is not the mechanism, whatever else is.

Arm three: a registry-based comparison, to size the effect at population level. Using adult diagnostic or insurance registry data of the kind Yerys and colleagues used for prevalence,¹⁰ compare age at first autism diagnosis between adults with a pre-existing ADHD diagnosis and those without, stratified by gender and by the calendar year of assessment, to separate a genuine cohort effect from a secular rise in diagnosis rates generally. This is the arm that would finally answer the plain version of the question: how many late-diagnosed adults are AuDHD, and is that proportion actually elevated relative to early-diagnosed adults, once cohort and gender are held constant?

Table 1
Three Accounts of the ADHD–Late-Diagnosis Link, and What Would Separate Them

Account	Where the effect lives	Predicted result in Arm One (vignette)	Predicted result in Arm Two (camouflaging battery)
Diagnostic overshadowing (including diagnostic sufficiency)	The clinician	ADHD label alone lowers referral likelihood, even among careful clinicians	No relationship - the patient hasn’t changed
Camouflaging / masking	The patient, compensatory	No effect of label alone	Camouflaging score predicts age at referral
Distinct developmental phenotype	The patient, constitutional	No effect of label alone	Camouflaging doesn’t mediate; genetic and neural evidence show a structurally distinct developmental trajectory789

Note: All three accounts predict the same raw correlation - AuDHD adults diagnosed later - which is why the correlation alone cannot choose between them. Only Arm One can isolate the clinician’s contribution; only Arm Two, combined with the genetic and neuroimaging work already published in children, can separate a compensatory strategy from a different underlying developmental course.

What I would actually do

• Run the vignette experiment first. It is the cheapest arm, requires no patient recruitment, and directly tests the mechanism - clinician behaviour - that is most immediately fixable if confirmed.
• Stratify every analysis by gender from the outset. Gesi and colleagues’ finding that men and women get overshadowed by different labels means a pooled analysis risks finding nothing because it is averaging over two distinct pathways.¹⁹
• Recruit the cognitive-camouflaging cohort at the point of referral, not after diagnosis - recruiting at referral - it is the only point at which the outcome is not yet known and cannot bias who enrols.
• Use the Camouflaging Autistic Traits Questionnaire and a cognitive-ability measure together, since Milner and colleagues found the camouflaging-diagnosis link differs by gender, and an unmeasured cognitive-ability difference could otherwise masquerade as a masking effect.¹⁷
• If resources allow, add a resting-state neuroimaging measure to Arm Two. Watanabe and colleagues’ pediatric finding - that the autism signature is neurally intact in AuDHD while the ADHD signature runs on different machinery - has never been tested in adults; replicating it in the same cohort recruited for the camouflaging battery would show directly whether the dissociation still holds once the population has actually reached the age this proposal is about.⁸
• Fix a definition of “late” before collecting a single data point, and justify it explicitly, given Russell and colleagues’ finding that the field currently has none.¹⁸
• In the registry arm, hold calendar year of assessment as a covariate, not an afterthought - diagnosis rates for adults generally are rising regardless of AuDHD status, and any AuDHD-specific effect has to be shown net of that background rise, not on top of an uncorrected one.
• Report the null result plainly if the vignette arm shows nothing. A finding that clinicians are not, in fact, swayed by a prior ADHD label would rule out overshadowing specifically, which narrows the explanation to masking or phenotype and is progress rather than failure.
• Design the vignettes and the battery with autistic and AuDHD adults, not just about them.

My honest prediction is that all three mechanisms will show something, in different proportions for men and women: overshadowing will show a real but moderate effect in the vignette arm - and it will look more like diagnostic sufficiency than diagnostic carelessness - camouflaging will mediate more of the effect in women than men, and a residual difference will remain even after both are accounted for - a residual that the Zhang et al. genetic trajectories and the Watanabe and Norman neural findings would predict, and that neither clinician training nor a better questionnaire could fix, because it would not be an error in anyone’s judgement. It would be a different course.

I could be wrong on all three counts. That is rather the point of doing the study.

Why this one is on the list

Because the three explanations point at three completely different remedies, and confusing them wastes the one that would actually work.

If overshadowing is the main driver, the fix is nearly free: a rule that a prior ADHD diagnosis cannot close the question of autism, built into referral pathways and clinician training. That is a checklist, not a trial - and the neural evidence suggests it is worth making that rule even for clinicians who are doing everything else right, since the omission this proposal is chasing does not require anyone to have erred.⁸⁹

It is worth being precise about why this checklist does not already exist, given that DSM-5 removed the old exclusion criterion in 2013. That change ran in only one direction: it stopped blocking an ADHD diagnosis in someone who already had autism - the mirror-image problem described above.¹⁵ It never touched the direction this proposal is about, because no rule ever blocked autism being diagnosed in someone who already had ADHD; there was nothing there for DSM-5 to remove. And permission to diagnose both is not the same as a habit of checking for both - the manual change created no prompt, requirement, or training update that would make a clinician keep looking once ADHD already explains the referral. The persistence of the delay in cohorts diagnosed well after 2013, including Diemer and Gerstein’s Missouri Medicaid sample above, is the evidence that the manual change and the clinical habit are two different things, and that only the second one is what this proposal is chasing.⁵

If masking is the main driver, the fix is a better instrument - something built to detect autism underneath a demonstrated capacity to compensate, rather than one that scores compensation itself as evidence against the diagnosis. That is slower and more expensive, but it is ordinary clinical science, not a new theory.

If a genuinely distinct developmental phenotype is the main driver - the Zhang et al. trajectory in which the difficulty is real but does not resemble early-childhood autism at all - then no amount of clinician training or better instrumentation solves it, because there was nothing early to catch.⁷ That would argue for treating late-emerging, ADHD-linked autism as its own recognised presentation with its own criteria, rather than continuing to measure it against a prototype built from three-year-olds it does not resemble.

Each of those is actionable. None of them requires new legislation, and none of them requires anyone’s permission to start. What they require is knowing which one is actually operating, in which people, and that is exactly the thing nobody has measured. An autistic adult who spent thirty years with an ADHD diagnosis and no autism diagnosis has usually been told, in one way or another, that the ADHD was the whole story. If a clinician stopped looking, that is worth knowing and worth fixing. If the presentation itself was different, that is worth knowing too - because it is not a story about being missed. It is a story about being a different shape than the one anyone was checking for, and nobody can explain that difference to somebody until the research says which one it actually was.

⁂

---