PDA scores
tracked ADHD
traits and
emotional
reactivity, not
autism, in a
UK clinic study
At one UK clinic (n=63), parent-rated pathological demand avoidance (PDA) didn’t differ by autism diagnosis but tracked ADHD traits and emotional reactivity.
Demand avoidance measured across a specialist clinic sample, not only in the autistic children
Pathological demand avoidance has spent most of its life inside autism. Elizabeth Newson described it in children referred to her clinic, and the profile has been argued over ever since: whether it belongs on the autism spectrum, whether it is better seen as a feature of several conditions, and whether ‘pathological’ is even the right word. Less often asked is the plain empirical question underneath: measure demand avoidance in a group of children, autistic and not, and what does it travel with?
A study published in Autism Research in 2026 asked exactly that. Oliver Lack, Francesca Happé, Emily Simonoff and colleagues at King’s College London and the Evelina London Children’s Hospital analysed parent-rated questionnaires for 63 children and young people referred for an autism assessment at a national specialist neurodevelopmental clinic between 2015 and 2018. The paper uses the term extreme demand avoidance (EDA), with “extreme/pathological” in its opening definition, and measures it with the Extreme Demand Avoidance Questionnaire (EDA-Q), a 26-item parent-rated measure built mainly from Newson’s original criteria.
The children were assessed by a multidisciplinary team against DSM-5 criteria, informed by the ADOS and the ADI-R. Thirty-one were diagnosed autistic and 32 were not. They were aged 6 to 18, with a mean age of 10, and 70% were male. Twenty-three, 36.51%, scored above the EDA-Q’s cut-off.
EDA-Q scores did not differ by autism diagnosis: a mean of 40.23 in the autistic children and 39.09 in the children who were not autistic, a difference that was not statistically significant (p = 0.78). Nor did scores differ by gender. And autistic traits, measured as a continuous score through the parent-reported autism module of a computerised diagnostic interview, did not correlate with EDA-Q scores either.
What did correlate was a different cluster. Across the whole sample, EDA-Q scores were negatively correlated with age and positively associated with emotional reactivity, ADHD traits and cognitive inflexibility. Weak associations were found with intolerance of uncertainty and with alexithymia, difficulty identifying and describing one’s own emotions. Anxiety did not correlate with EDA-Q scores.
This differs from the Directory’s earlier piece on the circular logic of a PDA diagnosis that can’t be validated, which was built on Company and Rotella’s 2026 systematic review of whether the tools used to identify PDA work. That piece was about the category; this one is about a single dataset. By the authors’ account it is also the first study to test alexithymia alongside EDA traits.
The data are a secondary analysis of anonymised clinical records already used in two earlier publications. The study had no direct funding; Lack’s PhD, the paper discloses, is funded by the De Pass Studentship, a philanthropic donation to King’s College London “to support two students researching PDA”.
Emotional reactivity, ADHD traits and age: what predicted the scores once the predictors were held against each other
The candidate drivers are tangled together: in this sample anxiety, intolerance of uncertainty, alexithymia and cognitive inflexibility were strongly associated with each other. So the authors built path models, which ask which predictors still carry weight once their overlap is accounted for.
Anxiety and autistic traits were left out of the models, because neither had correlated with the EDA-Q. Five predictors went in: ADHD traits, cognitive inflexibility (difficulty shifting between tasks or contexts, which the authors describe as a domain of executive function), emotional reactivity, intolerance of uncertainty and alexithymia, with age as a covariate. In the full model, only two were significant: emotional reactivity, from the Emotion Dysregulation Inventory, with a standardised coefficient of 0.46 (p < 0.001), and ADHD traits, from the hyperactivity scale of the Strengths and Difficulties Questionnaire, at 0.23 (p = 0.02). Age, at −0.19, was “just above the 0.05 threshold” (p = 0.051). Cognitive inflexibility (0.17), alexithymia (−0.05) and intolerance of uncertainty (−0.04) did not reach significance.
With the non-significant paths removed, in the final model emotional reactivity (0.52), ADHD traits (0.25) and age (−0.21) were the significant predictors, and together the model explained 54% of the variance in EDA-Q scores (52% after adjustment for over-optimisation). The abstract puts it in one line: “Only emotional dysregulation, ADHD traits and age predicted EDA-Q scores.”
| Predictor | Full model β | Final model β |
|---|---|---|
| Emotional reactivity (EDI reactivity) | 0.46 | 0.52 |
| ADHD traits (SDQ hyperactivity) | 0.23 | 0.25 |
| Age | −0.19 (p = 0.051) | −0.21 |
| Cognitive inflexibility | 0.17 (not significant) | removed |
| Alexithymia | −0.05 (not significant) | removed |
| Intolerance of uncertainty | −0.04 (not significant) | removed |
Source: Lack et al. (2026), Autism Research, Table 2. Final model explained 54% of the variance in EDA-Q scores (adjusted 52%). Exploratory analysis; no correction for multiple tests.
Two points are easy to lose. The first is the dropped predictors. Intolerance of uncertainty and alexithymia did correlate with demand avoidance; they stopped predicting it once emotional reactivity and ADHD traits were in the model. The authors suggest the predictors may share too much variance to separate, or that emotional reactivity sits closer to the behaviour, “thus masking associations with other predictors, which may be more distal”. They float a chain in which emotional dysregulation mediates between cognitive inflexibility and demand avoidance, a hypothesis the sample was too small to test.
The second is anxiety, which has been proposed as a central driver of demand avoidance: avoiding a demand as a way to escape the anxiety it provokes. Here there was no association, and the authors write that “This lack of an association is contrary to current conceptualizations of EDA”. They also say why it might be a measurement issue rather than a true absence: only normative T-scores were available from the anxiety measure, the RCADS, and they were skewed, with several children near the top of the range. Transformations and non-parametric tests made no difference, and the authors suggest the high scores mean anxiety was not underestimated. They do not claim the question closed.
On autistic traits, the paper reports earlier studies as mixed: O’Nions et al. (2014), White et al. (2023) and Rai et al. (2026) found associations with EDA traits, Rai et al. finding autistic traits the strongest statistical predictor over anxiety and intolerance of uncertainty; O’Nions et al. (2021) did not. The ADHD link has precedent too. As Lack et al. report it, O’Nions et al. (2021) found associations between SDQ-measured ADHD traits and EDA-Q scores, and Egan et al. (2020) a strong association with EDA self-ratings in adults, while McFadzen (2020) found no EDA-Q difference between children with and without an ADHD diagnosis. The authors call their own results on autistic traits and anxiety “surprising findings” that “do not support previous literature”.
Same questions, different tool: how far the measure may be finding its own definition
This paper is unusually candid about its instruments. The EDA-Q was largely derived from Newson’s original criteria, alongside the EDA items in the DISCO clinical screening tool and expert clinical consensus. Newson’s criteria included “a lability in mood led by the need to be in control”. The strongest predictor in this study, emotional reactivity, measures negative mood that escalates quickly and is poorly controlled. The authors say so: emotional dysregulation is “a distinct construct from EDA”, but “there is conceptual overlap”, which “may also explain the strong associations found in this study”. Part of the strongest result may be two questionnaires asking parents about related behaviour in different words. Then comes the sentence that makes the paper worth reading closely.
It is possible that use of the EDA-Q in the current study introduces further circularity into the literature. However, the measure is the only parental report EDA trait measure available at the current time, and we believe the current findings provide an important advance in empirical research on this topic.
That is the corpus’s oldest thread, stated by the researchers themselves: the instrument encodes the verdict. A questionnaire derived from one clinician’s description will, to some degree, find the features that description bundled together. The authors’ answer is a demand avoidance measure developed independently of Newson’s criteria. Until one exists, most work in this field measures demand avoidance with a ruler cut to the shape of the thing it is measuring.
The ADHD result has an instrument problem of its own. ‘ADHD traits’ here means the hyperactivity scale of a general screening questionnaire, scored 0 to 10. The sample’s mean was 8.1 and its median 9, and the scores were skewed, with several children near the top of the range. The authors suggest “more extensive ADHD trait measures containing a greater variety of items”. A crude measure still predicted demand avoidance, but it cannot say what about ADHD is doing the predicting.
And every measure came from parents. The authors note that parent-report questionnaires produce shared method variance, inflating associations between measures, and that parental report has been suggested to be more accurate for children’s behaviour than for their internal or emotional states. That bears hard on constructs like alexithymia and anxiety. A related problem turned up in a different population, in a Bristol study of young adults the Directory covered, where self-reported ADHD and autism trait scores correlated with each other but not with lab tests of executive function. Questionnaires agree with questionnaires.
A profile filed under autism, scored across a clinic sample
What this dataset adds is demand avoidance measured dimensionally across a clinical sample, rather than sought only in autistic children. A 2026 scoping review of nine schooling studies, by Áine Steenson, Nichola Booth and Audrey Roulston at Queen’s University Belfast, notes that “Much of the research within this review reflects the assumption that PDA is a part of the autistic spectrum”. Its own position is that “This review supports a strong link between PDA and autism, however, a range of other neurodivergent conditions were also reported”. ADHD, it says, “was strongly featured”, citing three of the nine studies. As the review reports those studies: in Reilly et al. (2014), all four children with a PDA profile met criteria for ADHD and three met criteria for autism; in Truman et al. (2024), almost twice as many children with autism and a PDA diagnosis also had an ADHD diagnosis, compared with autistic children without the profile; and in Connolly et al. (2023), 57.4% of a school distress group presenting with a PDA profile had an ADHD diagnosis, and 85.6% were autistic.
The review also names the obvious counter-argument: children in other neurodivergent groups who present with a PDA profile may “also have undiagnosed autism”, missed “because of factors such as surface sociability and masking”. Masking is exactly what a single informant in a single setting can miss. But this clinic’s design makes that a harder explanation to reach for. The non-autistic children were not a community sample. They were referred to a national specialist service for an autism assessment, assessed with what the authors call “gold standard” measures, and not diagnosed. That does not rule out missed autism, but it is not a sample recruited from those who already identify with the profile.
This study does not say pathological demand avoidance is ADHD, or that ADHD causes it. It found that, in one clinic’s sample, a parent-rated hyperactivity score predicted a parent-rated demand avoidance score, alongside a stronger predictor, emotional reactivity, and a negative one, age. The authors’ framing is that EDA traits may exist “trans-diagnostically”, and that the interaction between autistic, ADHD and EDA traits “may also be an important direction for future research”.
The book’s framework of universal cognitive mechanisms is useful here, provided nobody mistakes it for this paper’s finding. ADHD and autism, on that argument, are not two kinds of brain but names for positions far out along shared axes everyone runs, which is how a person crosses both thresholds and is AuDHD. Read that way, a profile found on both sides of a diagnostic line is what you would expect if it tracks a configuration rather than a label. The Directory has covered a review examining executive function and emotion regulation as proposed transdiagnostic mechanisms across ADHD and autism. This study is compatible with that reading, and cannot prove it.
Where the demand comes from: the half of the equation no questionnaire scores
Every instrument in this study measures the child. Nothing in the dataset measures the demands.
The authors come closest in discussing their strongest result. Emotional dysregulation in neurodivergent populations, they write, can result from contextual factors and environmental triggers such as sensory overwhelm, and may have a functional purpose. Then: “It is possible that the emotional dysregulation captured in this study is the result of other unmeasured contextual/internal variables such as stigma, school difficulties or sensory sensitivities. Alternatively, EDA might be a behavioral response to the distress that arises from these contextual/internal factors, resulting in dysregulation.” Both start the chain outside the questionnaire. Neither was measured.
The schooling review shows what that outside can look like. It suggests that “a substantial number of children and young people who present with the profile are experiencing extreme distress within their schooling environment”, contributing to exclusion and poor mental health. As it reports Gore Langton and Frederickson (2016), 29% of children had “often” been unofficially excluded and 62% had at least one school placement change due to their educational needs. Across the studies it reviews, traditional behavioural strategies such as rewards and consequences were reported as ineffective. Its own hedges apply: no review protocol, mostly self-selecting samples, commonly small ones, and few comparison studies with control groups. The Directory’s piece on exclusion among autistic pupils in the Millennium Cohort pointed the same way, leaving the unmeasured school environment as the larger question.
The informant problem returns here. In Brede et al. (2017), the review reports a split in EDA-Q ratings “where none of the teachers rated children as having clinically significant PDA, whilst all parents did”. It offers parental bias or masking in school as explanations. Either way, the same children scored differently depending on who was asked and where they saw them. If masking explains it, the score is at least in part a score about the setting; if parental bias does, it is a score about the informant.
This is the demand structure as a variable. The behaviour the EDA-Q captures is real; parents are not inventing it. But it is a meeting point between a child’s configuration and a set of imposed demands, and a questionnaire that scores only one side of the meeting will locate all of it in the child. If demand avoidance tracks emotional reactivity and hyperactivity, as it did in this clinic, the useful question is less “what does this child have” than “what is this child being asked to do, at what pace, under whose control”. This study was not designed to answer it.
For families, the practical reading is narrower than any headline. Demand avoidance does not need an autism diagnosis to be real or to deserve support, and this study is a reason not to assume one. An assessment asking only about autism may be asking too narrow a question; the guides to getting an autism diagnosis and getting an ADHD diagnosis set out the routes. The assessments category lists providers. For school, the guide to SEND support for families covers advocates and educational psychologists; for home, see support for parents of neurodivergent children. Schools can find providers under education.
The structural problem is bigger than any one family’s route. Pathological demand avoidance has been studied mostly inside autism, measured mostly with an instrument derived largely from one description of it, rated mostly by parents, and theorised mostly as anxiety. This exploratory study, with all its qualifiers, found the scores level across the autism line and pointing at emotional reactivity and ADHD traits instead. The authors’ prescription is right: a measure that does not presuppose the profile, wider samples, children followed over time. Add one more, which the Belfast review, citing Kildahl et al. (2021), names as a gap: “The specific demands producing challenging behaviour have not been investigated within the literature”. Measure the demand. Until a study does, the field will keep finding the cause of demand avoidance in the only place its instruments are pointed: the child.
Citations
Lack, O., Hollocks, M. J., Baird, G. et al. (2026) — Psychological and Neurodevelopmental Correlates of Extreme Demand Avoidance in a Clinical Sample, an Exploratory Study — Autism Research
Steenson, Á., Booth, N., Roulston, A. (2026) — Pathological demand avoidance and schooling experiences: A scoping review examining the experiences of children and their families — Research in Autism
News and updates.
Straight to your inbox.