Adult Psychiatric Morbidity Survey: Survey of Mental Health and Wellbeing, England, 2023/4

The Adult Psychiatric Morbidity Survey (APMS) covers both mental health and neurodevelopmental conditions, including autism spectrum disorder (henceforth autism). Autism prevalence estimates in APMS have been derived using a multi-stage assessment process since 2007. This approach combines a screening tool administered in the phase one survey interview (the 20-item version of the Autism Quotient (AQ) questionnaire) with a more detailed semi-structured assessment among a sub-sample of participants in a phase two interview. The semi-structured assessment used is the Autism Diagnostic Observation Schedule (ADOS), which is a standardised observational assessment widely used in clinical and research settings to support the identification of autism. Published studies have demonstrated good psychometric properties and diagnostic validity when the instrument is administered and scored by appropriately trained assessors (Lord et al. 2012b).

Over the past decade, autism has become a more prominent topic in public discourse and media coverage (Karaminis et al. 2023). During the same period, commissioning and provision of, and demand for NHS autism services, as well as the proportion of adults diagnosed with autism (NHS England 2026) and the proportion of adults self-reporting autism (Tromans et al. 2026) have all increased. 

After APMS 2023/4 autism estimates were published on 27 November 2025, further scrutiny of the results revealed inconsistencies that prompted a careful review of the data collection and analytical methods, as well as quality assurance checks of previous publications. This investigation revealed an analysis error affecting the originally published prevalence estimate, as well as substantial, unexplained, variation between interviewers in ADOS assessment outcomes to an extent that no reliable prevalence estimate could be derived. 

This chapter, therefore, takes a different approach from the standard APMS results chapters. Instead of presenting a new autism prevalence estimate for APMS 2023/4, it sets out the challenges of measuring autism in a population survey and how APMS has been designed to address them and describes how autism was assessed in the 2023/4 survey. It then explains the analysis error, the investigation that followed, and what the findings mean for prevalence estimation. It presents other autism-related measures from the 2023/4 survey as context and concludes with suggestions for how the issues could be avoided in future publications of APMS.

The aim of this chapter is to provide a transparent account of the issues identified, the investigations undertaken, and their implications for the interpretation of autism data collected in APMS 2023/4.

The issues identified do not affect published APMS 2007 and 2014 autism results, which remain valid experimental statistics. Separately, a clarification has been made to the description of the methods in the 2014 Appendix B to improve transparency and aid interpretation. It does not change the 2014 estimates.

The autism prevalence estimate published for APMS 2023/4 in November 2025 should not be used. It was produced from a variable that misclassified cases, and the corrected data are affected by the measurement issues set out in this chapter.

Background to autism

The concept of autism gained recognition in the mid-20th century, and its clinical definition continues to evolve over time (Frith 1991; Silberman 2015; Lord et al. 2020). Internationally recognised expert groups have achieved a broad consensus on what constitutes the category of autism, and diagnostic guidance set out in the fourth and fifth American Psychiatric Association (APA) Diagnostic and Statistical Manual (DSM-IV and -5) (APA 1994; 2013) and the World Health Organization (WHO) International Classification of Diseases (ICD-10 and ICD-11) (WHO 1993; 2024) are similar. Both systems require information on early childhood development and a substantial impact on functioning for diagnosis.

Being autistic can have an impact on learning, and, among those most profoundly affected, on the ability to live independently in adulthood (Howlin et al. 2004). Autistic adults often experience isolation and adverse experiences such as being bullied and socially excluded (Brugha et al. 2016). Furthermore, diagnosed autistic people have been found to have higher rates of a wide range of common mental and physical illnesses compared with the general population (Croen et al. 2015; Lai et al. 2019). 

The lifetime cost of supporting an individual diagnosed as autistic without intellectual disability was estimated in 2014 as £0.92 million in the United Kingdom, and higher for those with intellectual disability. Within these estimates, residential care, supported living accommodation and individual productivity loss contributed most to the lifetime costs for autistic adults with intellectual disability (Buescher et al. 2014). Quantifying such costs requires reliable estimates of the number of autistic adults, as well as the proportion who have a formal diagnosis. APMS 2007 was the first national general population probability survey to produce estimates using an examination for autism in a large sample of adults living in private households (Brugha et al. 2011). In APMS 2023/4, participants were, for the first time, also asked whether they thought they were autistic, and whether a professional had diagnosed them as autistic. The 2007 and 2014 APMS estimated the population prevalence of autism among adults living in households in England, to be about one in a hundred (Brugha et al. 2009; McManus et al. 2016). The 2021 Global Burden of Disease (GBD) study found a similar prevalence of about one in every 127 people, with an estimated 61.8 million autistic people globally in 2021 (Santomauro et al. 2025).

Reliable estimates of autism prevalence from general population surveys are important for understanding the number of autistic adults in the population, associated support needs, and future demand for services.

Challenges to measuring autism in population surveys

Autism is a highly complex neurodevelopmental condition, and scientific understanding of its scope and nature is still evolving. There is no single assessment instrument that serves as a definitive standard for diagnosing autism in adults. Instead, the National Institute for Health and Care Excellence (NICE) guidelines for diagnosing autism in adults recommend the use of a range of assessment tools as part of a comprehensive clinical assessment and emphasise that their role is in aiding, rather than replacing, clinical judgement (NICE 2012).

While clinical assessment of autism can be complex, it is even more challenging in the context of a general population survey such as the APMS. The challenges include:

• The need to use semi-structured assessments. Surveys typically measure most mental health and neurodevelopmental conditions by using a series of closed survey questions administered by survey interviewers. Given the complexity of autism and how it presents, it is not sufficient to use this approach. Instead, it requires a semi-structured assessment comparable to what is used in clinical practice. 
• It is not possible to fully assess all participants. Semi-structured autism assessments require more time than can typically be allocated within a general population survey that covers a range of mental health and neurodevelopmental conditions. The assessment also needs to be carried out by specially trained assessors whose time is more costly. This challenge can be dealt with by screening for autistic traits in all participants using a shorter questionnaire and conducting the semi-structured assessments with participants with a higher likelihood of being autistic in a separate follow-up interview carried out by specially trained interviewers. By not assessing all participants, some autism cases will invariably be missed.
• The need for a screening tool. The two-stage approach described above requires a reliable and sufficiently validated screening tool, to ensure that as few as possible participants with autism are screened out and as few as possible participants without autism are screened in.
• A clinical diagnostic assessment cannot be fully replicated. While it is possible to incorporate a semi-structured assessment in phase two of a general population survey, it is not feasible to fully replicate a clinical assessment. A clinician would make a diagnosis drawing in from a wider range of sources than a single assessment tool. In the case of diagnosis of autism, information from people who can report on a participant’s childhood experiences is ideally required, which is usually not feasible in the context of a population survey.
• Reliability of semi-structured assessments can be low. Semi-structured assessments are inevitably more complex and more aligned to professional judgement than simple survey questions and can therefore have lower reliability. Research using ADOS has shown that achieving reliability can be difficult and that ratings can differ across research centres (Lord et al. 2012a). Research also shows that reliability increases with assessor experience and can be improved with careful training and monitoring (Kamp-Becker et al. 2018). 
• Assessments not carried out by clinicians. Given the size and budgets of population surveys, it is not feasible for assessments to be carried out by clinicians experienced in diagnosing autism. Instead, research interviewers can be recruited to carry out assessments. While they receive in-depth training in the diagnostic tools used, it is not equivalent to the years of training and experience that a clinician will often have. 
• Maintaining comparability over time. Once an approach to autism measurement is established in a repeat survey, it cannot be changed in future surveys in the same series without affecting the time series. This means that new developments in clinical autism assessment or further understanding of chosen tools cannot be incorporated without replicating the validation process to understand the impact on timeseries. 

The approach to measuring autism in the APMS series was designed to address all these challenges, as outlined in the following section.

Case assessment of autism

In surveys of autism in childhood, information on behaviour and development is collected from parents and teachers. For adults, the most robust method of assessment would involve direct observation of current behaviour and information on both early childhood development and current day-to-day functioning over an extended period. This is not a practical option for a large general population survey of adults. Therefore, the assessment process used on APMS 2007, and replicated in APMS 2014 and 2023/4, was based on a combination of self-report screening questionnaire data collected at the phase one interview and a semi-structured examination carried out by a specially trained research interviewer at the phase two interview (Brugha et al. 2009). This multi-stage case assessment for autism is similar in structure to that used in the APMS series since 1993 for the assessment of psychotic disorders. This work was carried out by a consortium of research teams at the National Centre for Social Research and the University of Leicester, working jointly since the 2007 APMS.

The process involved the following stages: 

A. Phase one AQ-20 self-completion questionnaire 

B. Selection of cases for phase two assessment 

C. Phase two ADOS examination of a subset of selected adults 

D. Weighting to adjust for selection probabilities and non-response. 

This original approach has undergone an extensive programme of validation work at the University of Leicester, supported by NHS England and the Department of Health and Social Care (Brugha et al. 2009; 2011; 2012). The validation programme involved calibration of the ADOS with other research instruments for autism assessment including the Autism Diagnostic Interview-Revised (ADI-R) (Lord et al. 1994) and the Diagnostic Interview for Social and Communication Disorders (DISCO) (Wing et al. 2002); interviews with participants’ parents and other family members; comparison with further data collected from community; consensus ratings of APMS participant vignettes by autism practitioners; and engagement with psychiatrists and epidemiologists with expertise in this field. 

A. Phase one interview: Autism Spectrum Quotient questionnaire

The full Autism Spectrum Quotient (AQ-50) was one of few fully structured questionnaires designed to screen for possible autism in adults available at the time of the 2007 APMS. The AQ-50 had been found in clinical populations to have good correspondence with a full autism diagnosis (Baron-Cohen et al. 2001). Other available validated questionnaires tend to be longer (Ritvo et al. 2008) or require a parental informant that is not feasible in an adult household survey. A clinical diagnosis cannot be derived from the AQ-50, it is a test designed to identify potential underlying autistic traits of potential value in determining need for a clinical examination.

The full AQ consists of 50 items, covering five broad psychological dimensions: social functioning, imagination, communication, attention switching, and attention to detail. To minimise participant burden on the already long APMS 2007 questionnaire, a shorter 20 item version was derived using data collected by the original AQ developers and provided to the University of Leicester. Details of the modelling undertaken to select the best subset of items are given in a separate technical report (Brugha et al. 2011). The 20 items selected by the modelling procedure as the best predictors of a positive autism assessment spread quite evenly across these dimensions: six social functioning items; four communication; four attention to detail; three attention switching; and three imagination. The AQ-20 was discussed by an expert panel and tested in the cognitive pilot conducted as part of the APMS 2007 development work. Further modelling took place using a random sample of adults in contact with mental health services. This identified the 17 most predictive AQ items used in the 2007 survey, these were retained in subsequent publications of APMS and three (which had performed poorly in the 2007 survey data) were replaced with alternative items from the original AQ-50 McManus et al. 2016). The revised 20 item version of the AQ is reproduced in full in the questionnaire in Appendix C of the APMS 2023/4 Methods documentation. 

The AQ-20 is a self-completion questionnaire and was administered via a laptop in the phase one interview. 20 items were asked in each survey, including in 2023/4, but the score used to select participants for phase two, and to compare across survey years, was based only on the 17 items common to the 2007, 2014, and 2023/4 surveys. The remaining three items were not included in the score. Each response indicative of possible autism was given one point, giving a maximum score of 17, so that a higher score indicated greater likelihood that the person may be autistic. Because the AQ is a brief screening test and not a diagnostic measure, a clinical examination was included in the phase two interview. On APMS, the AQ was used to exclude participants with an extremely low likelihood of being autistic (those with an AQ score of between zero and three) and to inform the selection probabilities for phase two. It was not used to positively identify autism.

B. Selection of cases for phase two assessment

Participants were selected for phase two based on their phase one responses and agreement to be contacted by the University of Leicester, who conducted the phase two examination. In APMS 2014, the selection probabilities were based on participants’ AQ score and whether they were male or female. In APMS 2023/4, the same autism sampling fractions were applied to both male and female participants. In summary: 

• No participants with an AQ score of 0 to 3 were selected for a phase two ADOS examination on the basis of their AQ score. 
• A subsample of participants (16%) with an AQ score between 4 and 7 were selected for phase two. 
• All participants with an AQ score of 8 or more were selected for phase two.

It was not feasible for all phase one participants to have a phase two assessment, and this approach was designed so that those predicted to have a higher likelihood of being autistic had a higher likelihood of being assessed, combined with having a large enough sample size to be able to generate estimated rates of autism for the population as a whole.

C. Phase two assessment: Autism Diagnostic Observation Schedule (ADOS)

The second phase examinations were carried out by research interviewers from the University of Leicester. The Autism Diagnostic Observation Schedule (ADOS), Module 4, was completed with 884 participants at the APMS 2023/4 phase two; 628 in APMS 2014, and 618 in 2007. The ADOS is a semi-structured clinical examination of whether current behaviours are consistent with autism. It is a widely recommended ‘gold standard’ clinical research examination instrument for autism that is used to rate information on adult behaviour, that does not require involvement of an informant, which would have been impractical in an adult household survey (Lord et al. 2012b). 

The ADOS is made up of a series of standardised activities and conversational tasks that provide opportunities to observe communication, reciprocal social interaction, imagination, and stereotyped behaviours and restricted interests. In Module 4, which is used with verbally fluent adults, these tasks are largely conversation based. The examiner both delivers the tasks in a prescribed way and rates the behaviour observed, so the assessment relies on consistent delivery of a standard procedure and on consistent judgement in rating what is seen. 

As the ADOS does not offer sufficient opportunity to measure restricted and repetitive behaviours, which are emphasised in DSM-5 and ICD-11, additional examination items on restricted, repetitive patterns of behaviour and sensory differences were included in the phase two Schedule for Clinical Assessment in Neuropsychiatry (SCAN) interview, alongside a new set of neurodevelopmental items. The latter cover core ADHD symptoms and the components of autism required for diagnoses according to the DSM-5 (APA 2013) and the ICD-11. These new SCAN data have yet to undergo analysis and therefore will form part of future outputs not included in this chapter.  

The ADOS is one of the most extensively validated autism assessment instruments and is widely used in research and clinical practice internationally (Bastiaansen et al. 2011; Hus & Lord 2014; Wigham et al. 2019). Most of this evidence relates to the modules used with children, while Module 4, which is used with adults, has been studied less. The ADOS and its published clinical algorithm have been validated in previous clinic-based testing, but prior to APMS 2007 it had rarely been used with older adults or in a general population setting (Gotham et al. 2008). Determining ‘what is a case’ (the case threshold) in a population survey and making a diagnosis in clinical practice with patients seeking a clinical assessment should not be presumed to be the same. Therefore, methods and results of a quality assurance and validation study made use of experienced clinician ratings of survey data and of standardised developmental interviews conducted with parents and other informants to guide severity and clinical significance thresholds (Brugha et al. 2009). That study found that the ADOS performed well, and its results have informed the use of a case threshold of 10 or more. This was used in APMS 2007, 2014 and 2023/4. 

Assessors rated individual items on a scale from 0 to 3, with higher scores indicating behaviour more consistent with autism. The items covered two domains: communication and social interaction, and were summed to produce an overall score. Participants reaching the case threshold of 10 were classified as autism positive. As in APMS 2007, the score was calculated using the original item ratings. In clinical use of the ADOS, item scores of 3 are typically recoded as 2, but the APMS approach was originally validated without this step and it has therefore not been applied in any APMS publications. The only recoding applied was of scores of 8 (indicating a missing value), which were recoded to 0, consistent with the official ADOS algorithm (Lord et al. 2012b). Domain specific ADOS thresholds, which require a minimum score in each of these two domains as well as a total of 10 or more, were not applied. Both these decisions were taken to ensure that the definition of testing ADOS positive remained consistent across the survey series. Adults with a phase one AQ score of 3 or less were designated autism negative. This designation took precedence over the phase two assessment, so it applied even in cases where an ADOS examination had been completed for participants who did not meet autism selection criteria in phase one but did meet psychosis and/or eating disorder selection criteria.

D. Weighting to adjust for selection probabilities and non-response

Separate autism weights were calculated for analysing autism data as follows: 

• For participants who were both eligible for and took part in phase two, assessment weights adjusted for their probability of being selected for phase two interview for autism and for responding in phase two;
• Participants who were eligible for being selected for an autism assessment based on their AQ score but who were not selected and those who were selected but did not take part were treated as non-responders and did not receive weights;
• Phase one weight was used for participants who were not eligible for phase two.

The sampling and weighting strategy is described in more detail in Section 4 Weighting the data in the APMS 2023/4 Methods documentation.

Interviewer recruitment and training

For APMS 2000, 2007 and 2014, all recruited phase two interviewers had an undergraduate or graduate psychology degree. For 2023/4, a psychology degree was considered a desirable rather than essential criterion due to changes in University of Leicester inclusivity regulations regarding the appointment process. Nonetheless, all recruited interviewers were graduates and/or had clinical experience in a medical, social science or related field.

Following advertisement of the role, all interviewers were selected via CV review and an interview involving a panel of University of Leicester staff experienced in psychology, psychiatry, and epidemiological research. None of the phase two interviewers in APMS 2023/4 had previously worked on APMS 2007 or 2014 (similarly, there were no interviewers who had worked on both APMS 2007 and 2014). The APMS 2023/4 training and fieldwork manager worked across both the 2014 and 2023/4 surveys and carried out some interviews in 2014.

All interviewers in 2007, 2014 and 2023/4 underwent an ADOS training course at the University of Leicester, delivered by an accredited international ADOS trainer with over two decades of experience (Fiona Gullon-Scott). Inter-rater reliability was assessed on multiple occasions during the one-month training course using recorded ADOS interviews, with ratings compared and discussed to support alignment across interviewers. Consistent with standard requirements for research reliability, a minimum inter-rater reliability threshold of 80% was applied. Interviewers met this criterion during training. During fieldwork, the training and fieldwork manager accompanied each interviewer on three assessments and rated these independently. Agreement remained at or above the 80% threshold for every interviewer. Ratings were also discussed at weekly team meetings, although these did not involve re-rating of assessments carried out with participants.

Following the ADOS training course, all interviewers received ongoing support and quality assurance during fieldwork. Weekly team meetings were held to review interviews, discuss fieldwork progress, and address any uncertainties in ratings and refine practice where required. As discussed above, the training and fieldwork manager observed each interviewer on at least three occasions to monitor adherence to training protocols and identify any potential drift in both practice and scoring. No major concerns were identified during these observations. Only minor adjustments were noted, for example, optimising the placement of ADOS materials during activities.

All training procedures, ADOS administration, and phase two fieldwork management were conducted in full accordance with the protocols established in 2014. Those procedures were initially developed for the APMS 2000 phase two and further refined in the 2007 APMS in which autism was first included.

Differences from previous APMS

APMS 2023/4 was designed to be as consistent with previous APMS surveys as possible to ensure the results were comparable and trend analysis could be conducted. All design and implementation differences are described in the APMS 2023/4 Methods documentation. The ones potentially relevant for autism measurement are:

• Lower response to phase one interview. The response rate achieved in APMS 2023/4 (29.4%) was lower than previous surveys in the series (57% in 2014), consistent with broader trends in survey non-response rates (de Leeuw, Hox & Luiten 2018; Williams & Brick 2018). Phase one data were weighted to take account of non-response and to ensure that the sample matched known population totals. It is, however, possible that some non-response bias related to autism remained in the data. While lower response rates may increase the risk of non-response bias, evidence suggests that response rate alone is a poor predictor of its magnitude (Groves & Peytcheva 2008).
• Inclusion of eating disorders in phase two interview. A SCAN eating disorder assessment was included for the first time in APMS 2023/4. This meant that more participants were eligible for phase two, and that the phase two assessment took longer. Because eligibility was widened in this way, the mental health profile of phase two participants in 2023/4 differed from that in previous publications, and evidence suggests that differences of this kind can influence ADOS ratings (Conner et al. 2019). The extent of any effect on the APMS 2023/4 data could not be established, although a separate systematic review has reported high co-occurrence between eating disorders and autism (Nickel et al. 2019).
• Inclusion of SCAN autism assessment in phase two interview. Also, for the first time, phase two assessment included a recently developed SCAN autism assessment. Whilst it is not expected that this change would have influenced how participants scored on ADOS, this cannot be ruled out.
• Profile of phase two interviewers. In all previous APMS surveys, phase two interviewers were required to have an undergraduate or graduate degree in psychology. This subject criterion was relaxed in 2023/4, which resulted in no phase two interviewers having psychology degrees.

 Identification of an analysis issue

Following publication of the APMS 2023/4 report in November 2025, secondary analysis using the APMS 2023/4 dataset identified an error in the final stage of deriving the binary autism classification variable used for reporting autism prevalence estimates. In the APMS analysis, ADOS scores of 10 or above were intended to indicate autism-positive cases. However, the classification threshold was applied incorrectly, such that cases with scores of 7 to 9 were classified as autism positive, while cases with scores of 10 or above were classified as autism negative.

The error only affected the reporting variable used to produce the published autism prevalence estimates for APMS 2023/4. As part of the subsequent investigation, the equivalent syntax from APMS 2007 and 2014 was reviewed and checked against the study methodology. No corresponding issues were identified, indicating that the analysis error was specific to APMS 2023/4.

The analysis error was corrected, and the analyses were re-run. Following correction, the resulting autism prevalence estimate was substantially higher than the estimate published in the original report and suggested an increase in autism prevalence relative to APMS 2014 that was implausible given the time interval between surveys and the wider evidence base. This unexpected finding prompted a wider investigation of the autism assessment data, analytical processes and underlying assumptions.

The findings of this further investigation are described in the following sections.

Further quality assurance checks

Following identification of the analysis error in the original chapter, a programme of additional quality assurance checks was undertaken to verify the derivation of the autism prevalence estimates and produce a corrected estimate. The purpose of this work was to establish whether any other issues were present that could affect the validity of the estimates and to ensure that the published analysis accurately reflected the underlying assessment data. This work was conducted jointly across consortium institutions and involved both methodological and clinical review.

Data pipeline and syntax checks

Consortium members independently carried out syntax and data processing checks to verify that participants were correctly routed through the phase one and phase two questionnaires, that responses were accurately transferred from the Computer Assisted Personal Interviewing (CAPI) system to the SPSS analysis dataset, and that derived variables had been generated as intended. This review involved tracing a purposive sample of participants through the assessment pathway and comparing questionnaire specifications, raw interview data and analysis datasets. No routing errors, data transfer issues, mismatches or omissions were identified, providing assurance that the observed findings were not attributable to errors in the questionnaire programming, interview routing, or the data processing pipeline.

Independent reviews of the autism variable derivation were also undertaken by three researchers from both consortium institutions. Each researcher independently reviewed the derivation syntax against the documented methodology and assessment scoring criteria before comparing findings. The reviews reached the same conclusions regarding the derivation of the autism variables, providing assurance that the analysis syntax was being implemented consistently and in accordance with the intended methodology.

Analysis of ADOS scoring patterns

Following correction of the analysis error, patterns in ADOS score and autism classification were reviewed to understand the drivers of the revised prevalence estimate. This included examination of the distribution of ADOS scores, the number of participants classified as autism positive, and variation in outcomes between phase two interviewers. 

Substantial differences were observed between phase two interviewers in ADOS scores and in rates of autism identification. One phase two interviewer, who completed only two interviews, was excluded from this analysis. Among the remaining six phase two interviewers in APMS 2023/4, the proportion of their completed interviews identified as autism cases ranged from 0.5% to 30%, a range of 29.5 percentage points, with an interquartile range of 8.3. By contrast, among the four phase two interviewers in APMS 2014, the equivalent proportions ranged from 0.7% to 2.6%, a range of 1.9 percentage points, with an interquartile range of 1.3.

Chi-square test was carried out to quantify the variation in the proportion of autism positive cases between the six interviewers. This showed that the autism identification rates varied between interviewers more than could be expected by chance alone. 

The analysis was repeated on 2014 data, and the between-interviewer variation in the autism identification was found to be much lower and within a range that could be expected by chance, indicating that variation of the kind observed in 2023/4 was not present in 2014. Interviewer information is no longer available for 2007 data, so the analysis could not be repeated for that publication.

Further analysis was carried out to investigate whether interviewers’ case load in 2023/4 varied by characteristics known to be related to the probability of having autism. No apparent differences were found in the AQ scores, sex and age distribution of participants by interviewer, showing that between-interviewer variation was not due to them having a different mix of participants to assess. While interviewers and participants were geographically distributed across England, the scale of the variation observed between interviewers suggested that geographical variation alone was unlikely to explain the findings.

Eating disorder and psychosis SCAN data was also reviewed for between-interviewer variability to determine whether variability extended beyond ADOS assessments. The variability of SCAN results for those two conditions was found to be within the expected range and, for psychosis, in line with 2014 results, indicating that the variability was specific to ADOS data.

These findings showed that there was significant between-interviewer variation in ADOS results in 2023/4, which was not present in earlier years or for other conditions. 

Review by clinical experts and statisticians

Findings from the quality assurance programme were reviewed jointly by consortium researchers, clinical experts and statisticians, including members of the University of Leicester, the National Centre for Social Research, and independent statistical advisors. Discussions considered the consistency of the findings across the various analyses, the implications of the observed scoring patterns, and potential explanations for the results. This multidisciplinary review informed the decision regarding the interpretation of the autism assessment data and the extent to which a robust prevalence estimate could be derived, presented in the next section.

Significant between-interviewer variability in ADOS assessments suggests that interviewers carried out the assessments differently from each other in 2023/4 and not in 2014, indicating that some of them may have diverged from the intended application of ADOS. This may be partly due to the different profiles of interviewers, discussed above, although the survey provides no direct evidence of it. 

These findings relate to how the ADOS was administered in APMS 2023/4, not to the instrument itself. The ADOS remains a well validated assessment, and nothing identified here calls into question its use in previous APMS publications or in population surveys more generally.

Recognition of autism in wider society has been expanding over the past decade. Frith (2026) has argued that the concept in practice has broadened to the point where the diagnostic threshold being applied has effectively been lowered, driven in part by cultural change rather than by change in the underlying population. However, other factors may have contributed, including improved recognition of autism in previously underdiagnosed groups who do access services (Zeidan et al. 2022). Phase two interviewers work within the same cultural context as the wider public, and because ADOS assessments require them to judge the presence and severity of behaviours relevant to autism, this changing understanding may have influenced how individual assessment items were interpreted and scored. Although, there is no evidence from APMS 2023/4 that would allow the verification of this hypothesis, it will be an important issue for future research to address, given the prominence of the ADOS in both clinical and research contexts.

There is no way to objectively assess how and to what extent interviewers diverged from the intended application of ADOS, as ADOS assessments were not recorded for later independent verification. Therefore, it was not feasible to undertake retrospective independent re-rating of assessments to evaluate or adjust for potential interviewer differences.

It is therefore not possible to produce a reliable and robust autism prevalence estimate based on data from APMS 2023/4.

Self-reported autism and professional diagnosis of autism

Participants were shown a list of mental health and neurodevelopmental conditions and asked whether they thought that they had ever had any of them. In APMS 2023/4, the list was expanded to include ‘autism spectrum disorder’. Those who reported that they thought that they were autistic were asked whether this had been diagnosed by a professional.  

Self-reported autism and professional diagnosis of autism, by sex

Overall, 3.4% of adults reported that they were autistic, and 1.2% of adults reported having received a professional diagnosis of autism. Men (4.2%) were more likely than women (2.6%) to report that they were autistic, but they were equally likely to report that they were diagnosed as autistic by a professional.

For more information: Table 10.1

Self-reported autism and professional diagnosis of autism, by co-occurring common mental health condition

Those with a common mental health condition (CMHC) (10.7%) were more likely to report that they were autistic than those without a CMHC (1.5%). Those with a CMHC (4.0%) were also more likely to report that they had been professionally diagnosed with autism than those without a CMHC (0.4%).

For more information: Table 10.2

Interpretation

These findings should not be interpreted as estimates of autism prevalence or diagnosis. It is unknown what proportion of participants who reported that they either were or were not autistic would be identified as autistic on clinical examination, as robust assessment of autism requires a comprehensive, standardised clinical evaluation conducted by appropriately trained assessors. They are also based on self-report rather than administrative or clinical diagnosis data. 

Participants who had a professional diagnosis come closer to that standard, because those participants would have undergone a comprehensive clinical assessment. However, this approach excludes people who may be autistic but have not received a formal diagnosis, as well as people who would meet diagnostic criteria but do not identify themselves as autistic. 

Trend analysis of self-reported autism is not possible with APMS data because it was only included in APMS 2023/4, but an analysis of GP Patient Survey data showed that the proportion of adults saying they were autistic was 2.3% (95% CI 2.3, 2.4) in 2024, which is a significant increase compared with 2018 (0.9%, CI 0.9, 0.9) (Tromans et  al. 2026).

Autism Quotient (AQ) questionnaire

What the AQ-20 measures

The Autism Spectrum Quotient (AQ) is included in the APMS phase one interview to screen participants for the phase two interview. It is not intended as a measure of autism, and it has never previously been reported on in the APMS report series. The AQ findings below provide context on how prevalence of autistic traits has changed over time; they are not a substitute for the ADOS.

The AQ-20 is described in the section Case assessment of autism. When interpreting results based on the AQ, it is important to recognise that the AQ is not a diagnostic instrument (Baron-Cohen et al. 2001; Yoshinaga et al. 2023). Developed to quantify autistic traits in adults, the AQ has been approved for identifying individuals who may warrant further assessment and cannot, on its own, establish whether a person meets clinical diagnostic criteria for autism (Baron-Cohen et al. 2001; NICE 2012; Ruzich et al. 2015). 

The traits that the AQ measures do not all neatly map onto the diagnostic criteria in DSM-5 or ICD-11. As described in Section 10.3, the items were selected because they best predicted a positive phase two assessment.

There is ongoing debate regarding exactly what the AQ measures, with some evidence suggesting that its factor structure and construct validity are not straightforward, and that it may capture a broader range of social, cognitive, and behavioural characteristics rather than autism-specific traits alone (Lundqvist & Lindner 2017). Higher AQ scores have been shown to be associated with symptoms of common mental health conditions, including anxiety and depression, both in autistic and general population samples, raising the possibility that AQ scores may in part reflect broader psychological distress or related psychosocial difficulties rather than being specific to autism alone (Hollocks et al. 2019). Consequently, AQ-based estimates should be interpreted with caution and not treated as equivalent to survey-based estimates of autism prevalence required for national policy and planning.

AQ score of 8 or over in 2007, 2014 and 2023/4, by sex

The proportion of adults scoring 8 or over on the AQ in 2023/4 was 20.4% (95% CI 19.0, 21.8). This was an increase from 14.5% (CI 13.6, 15.4) in 2014 and 10.9% (CI 10.0, 11.8) in 2007. This rise was evident in both men and women and the proportion of men scoring 8 or over remained consistently higher than women across the survey series. 

For more information: Table 10.3

AQ score of 8 or over in 2007, 2014 and 2023/4, by co-occurring common mental health condition

Across all survey years, adults with a common mental health condition (CMHC) were more likely to score 8 or over on the AQ than those without a CMHC. 

Among men with a CMHC, the proportion scoring 8 or over rose from 25.7% in 2007 to 51.9% in 2023/4. A similar pattern was observed among women. In women with a CMHC, the proportion scoring 8 or over increased from 20.0% in 2007 to 35.8% in 2023/4.

The proportion of adults scoring 8 or over on the AQ increased over time among both those with and without a CMHC, so the overall rise cannot be explained by the increase in CMHCs alone.

For more information: Table 10.4

Interpretation

As mentioned above, the AQ is designed to measure potentially autistic traits in adults but may also be capturing broader psychological distress. APMS findings presented here show that adults who have a common mental health condition are more likely to score 8 or over on the AQ. This has been the case for all three APMS publications for which comparable data exists. At the same time, the proportion of adults with an AQ score of 8 or over as well as those with a common mental health condition has increased during this period. While these findings show a strong link between the two measures, they do not allow us to disentangle the exact relationship between them and, in particular,to shed further light on what the AQ is measuring. Notably, the proportion scoring 8 or over on the AQ rose from 10.9% in 2007 to 14.5% in 2014, while the ADOS based prevalence estimate remained at about one in a hundred in both years. The AQ therefore moved independently of the measure it was screening for, which reinforces the point that it should not be understood as a measure of autism prevalence in the general population.

The increase of the proportion of participants who scored 8 or over on the AQ in APMS 2023/4 meant that a larger proportion of participants were eligible for phase two assessment. This, together with the poor correspondence between the AQ and the ADOS observed when the approach was developed (Brugha et al. 2012) and its substantial overlap with common mental health conditions, shows that further work is needed to understand how the AQ can be used as a screening tool in future general population surveys. The AQ was retained in APMS 2023/4 because it was the most appropriate measure available when APMS 2007 was designed and because its continued use supported comparability across survey publications.

Studies reporting autism prevalence based on administrative records (Russell et al. 2022), as well as those based on self-reporting autism (Tromans et al. 2026) report an increasing trend over time. It is unclear whether these trends correspond to an increase in population prevalence, or are attributable to other factors, or both. General population surveys that assess autism directly in a random probability sample, such as APMS, are therefore a crucial source of evidence on how common autism is, the needs of autistic adults, and the demand for services. That is why the issues identified in APMS 2023/4 should be addressed to strengthen the approach for future surveys.

Further methodological work with APMS 2023/4

Further methodological work is planned by the University of Leicester with APMS 2023/4 data, which will shed further light on autism data and inform decisions about future changes to the autism measurement approach. The following analyses are planned:

• Analysis of SCAN autism data. Neurodevelopmental conditions have only recently been added to SCAN, allowing the SCAN autism module to be introduced to APMS in 2023/4 for the first time. We will analyse the SCAN data, including comparing it to other autism related data in the questionnaire and reviewing between-interviewer variability to establish if it is subject to similarly high variability as the ADOS.
• Further analysis of ADOS data. We will review ADOS data in detail to establish whether any elements are more prone to between-interviewer variability than others. 
• Further analysis of AQ data. We will review AQ data for its effectiveness as a screening tool and consider whether there are alternative screening tools that have emerged since the 2007 survey.

Alongside this, there is an option to extend APMS 2023/4 fieldwork with a third phase, to carry out repeat ADOS measurements on a subset of phase two participants, conducted by highly trained interviewers. The phase two assessments themselves cannot be re-rated because they were not recorded but repeat measurement would provide a comparison that the existing data does not offer. This approach was used in APMS 2007 to validate the current autism measurement approach. Such work could help determine the extent to which the findings reflect measurement issues, changes in the underlying population, or a combination of both.

Finally, future work should be designed to provide independent evidence of possible changes in autism prevalence estimates since 2014. This would give some indication of whether any observed changes are based on changes in the population or due to methodological differences. It is possible that changes in society since 2014 could contribute to increased ADOS scores, e.g. due to increasing screen time and reduced in-person social interaction among young people with their peers although evidence on this is inconclusive (Ophir et al. 2023). Future work could use age-period-cohort analyses to investigate whether any observed differences reflect ageing, changes over time affecting all age groups, or generational differences between birth cohorts.

Potential changes to autism measurement in future APMS surveys

There are several changes that may help to increase the quality of autism data in APMS. Some of these are evident from what we already know, whereas others depend on the findings from the methodological work described above. We are not proposing a change to the overall two-stage design, but the following adjustments should be considered within it, including whether the instruments used at each phase remain the right ones:

• Recruiting phase two interviewers who have an academic background in psychology. This was the most obvious change to the phase two model since 2014. There is no direct evidence for how it affected measurement, but it is one of the few identified differences that could bear on ADOS scoring. Reverting to the previous requirement should be considered, taking account of the recruitment policy reasons for the change described above.
• Video recording a subset of phase two interviews. As ADOS is observational, it was not possible to re-rate interviewers’ work after the event. A video recording of an interview would create an additional layer of quality assurance during fieldwork and allow for any re-rating should that be necessary.
• Maintaining consensus ratings beyond training. Interviewers were trained to a consensus standard, and each was accompanied and independently re-rated on three assessments during fieldwork, with agreement remaining above the threshold. This covered around 20 of the 884 assessments completed, too few to show how individual interviewers were scoring across their full caseload. Re-rating a larger proportion, spread across the fieldwork period, would give a firmer basis for identifying differences in scoring while fieldwork is still running.
• Data monitoring of the distribution of autism assessment scores during fieldwork. In 2023/4, as in 2007 and 2014, between-interviewer variability in autism scores was not assessed during fieldwork. The 2023/4 experience shows that reviewing scores periodically during fieldwork could have identified the variation while it was still possible to act on it.
• Consider whether AQ remains the optimal phase one interview screening tool. Should our analysis indicate that a more valid and reliable tool is available, then it should be subjected to methodological field testing ahead of the next APMS.
• Consider whether ADOS remains the optimal phase two assessment. If our proposed methodological work shows that SCAN is a more reliable measure of autism compared with ADOS, then we would need to consider whether it could be used in parallel with the ADOS in phase two interviews. This decision would need careful consideration and further field testing to ensure that we fully understand differences and therefore timeseries implications of the two measures.

All these initial suggestions would require further testing or methodological work to establish their impact on the quality of autism estimates. Furthermore, additional improvements may be identified through planned methodological work with APMS 2023/4 data.

All these adjustments also have cost implications, which would need to be considered when deciding whether to implement them.

Possible wider changes to APMS

The points in this section are separate from the issues identified in APMS 2023/4. They concern whether an instrument covering a wider range of conditions would serve the survey better in future, rather than explaining why the 2023/4 assessments varied between interviewers.

If SCAN is shown to be a strong alternative to ADOS, it opens an opportunity to consider more extensive changes to APMS to potentially improve measurement of a wider range of conditions. 

Condition specific instruments such as the ADOS only assess autistic features and provide relatively little guidance to examiners on how to differentiate these from features and symptoms associated with co-occurring mental health conditions and other neurodevelopmental conditions. For this reason, it is widely recognised that an ADOS assessment should be postponed if a person is acutely mentally ill, for example acutely depressed, because behaviour arising from that condition can be incorrectly scored as autistic as the ADOS has no other way of coding it (Conner et al. 2019; Maddox 2020). This may result in double counting, where the same behaviour contributes to a positive autism assessment and to a positive assessment on whichever separate instrument is used for the other condition, so the participant is recorded as having both. Measures that cover several conditions at once are more likely to be semi-structured with information being collected in a clinically flexible way. Examples used in surveys have included the Present State Examination (PSE) (Wing et al. 1974), the Structured Clinical Interview for DSM Disorders (SCID) (First et al. 2016) and the SCAN (World Health Organization, SCAN 3.0, in press).

The SCAN semi-structured interview not only covers psychosis, autism, ADHD, and eating disorders, but also anxiety and depressive disorders, use of psychoactive substances, alcohol and nicotine, in one assessment process. The SCAN is designed to differentiate across these different conditions and would potentially allow for more accurate measurement of not just autism but many other conditions. Work commencing at the University of Leicester with its World Health Organization partners is also examining the potential for artificial intelligence (AI) to be used in future approaches to the use of the SCAN.

We therefore recommend that the SCAN is shortlisted as a potential phase two measure for future APMS publications, because it is designed to differentiate between the psychiatric and neurodevelopmental conditions that matter most to the survey.

These are options for consideration rather than agreed changes. Which of them are taken forward, and in what form, will depend on the findings of the methodological work described above and, on their cost, and value. Further changes may also be identified as that work progresses. Future developments relating to the APMS, including associated academic outputs and publications, will be publicised through the webpages of NHS England, DHSC, the University of Leicester and NatCen.

The results described in this chapter show that a robust autism prevalence estimate cannot be produced from the APMS 2023/4 data. This is not because autism cannot be measured in a population survey. APMS 2007 and 2014 showed that it is possible, using a carefully designed two-stage process and a validated assessment approach. The issue in 2023/4 is that, once the analysis error had been corrected, the ADOS data showed substantial between-interviewer variation that could not be explained by the profile of participants assessed. Without an independent benchmark, and without recorded assessments that could be re-rated, there is no basis for adjusting the data after the event.

For that reason, this chapter does not present a revised autism prevalence estimate for APMS 2023/4. Presenting an estimate that is affected by unresolved measurement issues would risk giving a misleading account of autism prevalence in the adult population. The 2007 and 2014 prevalence estimates remain valid and robust, and the interviewer level variability identified in ADOS is specific to the 2023/4 reporting and assessment data.

The other autism-related measures collected in APMS 2023/4 are still useful, but they need to be interpreted carefully. Self-reported autism, professional diagnosis of autism and AQ scores provide context about changing awareness, recognition and reporting of autism. They also show the links between autistic traits, self-identification, diagnosis and common mental health conditions. However, none of these measures are a substitute for a robust clinical or semi-structured assessment, and none should be treated as an autism prevalence estimate.

The findings also point to possible improvements for future surveys. These include strengthening recruitment criteria for phase two interviewers, directly observing or recording more assessments so that ratings can be independently reviewed, monitoring interviewer-level scoring patterns during fieldwork, and reviewing whether the AQ remains the best screening tool. Further analysis of the new SCAN autism data will also be important in deciding whether it could be used for autism measurement instead of or alongside the ADOS. Any change would need careful testing, because maintaining comparability across APMS publications remains a primary aim of the series.

Although not currently planned, there would also be value in further follow-up work with APMS 2023/4 phase two participants, including repeat expert-administered assessments for a selected subset. This would not entirely address the limitations of the existing ADOS data. It could help to understand how far the observed findings reflect measurement issues, changes in the population, or both. 

More broadly the findings suggest that, as with all surveys, methodologies should be kept under review, and further work could inform development on autism measurement in adult population surveys. The aim now should be to learn from APMS 2023/4 and strengthen the approach for future surveys.

American Psychiatric Association (1994) Diagnostic and Statistical Manual of Mental Disorders, 4th Edition. 

American Psychiatric Association (2013) Diagnostic and Statistical Manual of Mental Disorders, 5th Edition. 

Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The autism-spectrum quotient (AQ): Evidence from Asperger syndrome/high-functioning autism, males and females, scientists and mathematicians. Journal of Autism and Developmental Disorders, 31(1), 5–17.

Bastiaansen, J. A., Meffert, H., Hein, S., Huizinga, P., Ketelaars, C., Pijnenborg, M., ... & De Bildt, A. (2011). Diagnosing autism spectrum disorders in adults: the use of Autism Diagnostic Observation Schedule (ADOS) module 4. Journal of Autism and Developmental Disorders, 41(9), 1256-1266.

Brugha, T. S., McManus, S., Bankart, J., Scott, F., Purdon, S., Smith, J., ... & Meltzer, H. (2011). Epidemiology of autism spectrum disorders in adults in the community in England. Archives of General Psychiatry, 68(5), 459-465. https://jamanetwork.com/journals/jamapsychiatry/fullarticle/211276

Brugha, T. S., McManus, S., Meltzer, H., Purdon, S., Scott, F., Baron-Cohen, S., ... & Bankart, J. (2009). Development and testing of methods for identifying cases of Autism Spectrum Disorder among adults in the Adult Psychiatric Morbidity Survey 2007. Leeds, UK: The NHS Information Centre. https://files.digital.nhs.uk/publicationimport/pub02xxx/pub02931/adul-psyc-morb-res-hou-sur-eng-2007-rep-v2.pdf

Brugha, T. S., McManus, S., Smith, J., Scott, F. J., Meltzer, H., Purdon, S., ... & Bankart, J. (2012). Validating two survey methods for identifying cases of autism spectrum disorder among adults in the community. Psychological Medicine, 42(3), 647-656. 

Brugha, T. S., Spiers, N., Bankart, J., Cooper, S. A., McManus, S., Scott, F. J., ... & Tyrer, F. (2016). Epidemiology of autism in adults across age groups and ability levels. The British Journal of Psychiatry, 209(6), 498-503. https://www.cambridge.org/core/journals/the-british-journal-of-psychiatry/article/epidemiology-of-autism-in-adults-across-age-groups-and-abilitylevels/D4F48E07D1002DE6F75A67FF9A4FFCF7

Buescher, A. V., Cidav, Z., Knapp, M., & Mandell, D. S. (2014). Costs of autism spectrum disorders in the United Kingdom and the United States. JAMA Pediatrics, 168(8), 721-728. https://jamanetwork.com/journals/jamapediatrics/fullarticle/1879723

Conner, C. M., Cramer, R. D., & McGonigle, J. J. (2019). Examining the diagnostic validity of autism measures among adults in an outpatient clinic sample. Autism in Adulthood, 1(1), 60–68. https://journals.sagepub.com/doi/10.1089/aut.2018.0023

Croen, L. A., Zerbo, O., Qian, Y., Massolo, M. L., Rich, S., Sidney, S., & Kripke, C. (2015). The health status of adults on the autism spectrum. Autism, 19(7), 814-823.

de Leeuw, E., Hox, J., & Luiten, A. (2018). International nonresponse trends across countries and years: An analysis of 36 years of labour force survey data. Survey Methods: Insights from the Field, 1-11. https://surveyinsights.org/?p=10452

First, M. B., Williams, J. B. W., Karg, R. S., & Spitzer, R. L. (2016). Structured Clinical Interview for DSM-5 Disorders, Clinician Version (SCID-5-CV). American Psychiatric Association Publishing.

Frith, U. (1991). Autism and Asperger Syndrome. Cambridge University Press: Cambridge.

Frith, U. (2026). Autism spectrum disorder: has it lost its meaning and is it leading to misdiagnosis? Psychological Medicine, 56, e240. https://www.cambridge.org/core/journals/psychological-medicine/article/autism-spectrum-disorder-has-it-lost-its-meaning-and-is-it-leading-to-misdiagnosis/AF834026D004D42B1963BEA515495AC6

Gotham, K., Risi, S., Dawson, G., Tager-Flusberg, H., Joseph, R., Carter, A., ... & Lord, C. (2008). A replication of the Autism Diagnostic Observation Schedule (ADOS) revised algorithms. Journal of the American Academy of Child & Adolescent Psychiatry, 47(6), 642-651.

Groves, R. M., & Peytcheva, E. (2008). The Impact of Nonresponse Rates on Nonresponse Bias: A Meta-Analysis. Public Opinion Quarterly, 72(2), 167–189. 

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(4), 559-572. https://www.cambridge.org/core/journals/psychological-medicine/article/anxiety-and-depression-in-adults-with-autism-spectrum-disorder-a-systematic-review-and-metaanalysis/CDC4FF29C3DC504768E375EE65019E0C

Howlin, P., Goode, S., Hutton, J., & Rutter, M. (2004). Adult outcome for children with autism. Journal of Child Psychology and Psychiatry, 45(2), 212-229. https://acamh.onlinelibrary.wiley.com/doi/full/10.1111/j.1469-7610.2004.00215.x

Hus, V., & Lord, C. (2014). The autism diagnostic observation schedule, module 4: Revised algorithm and standardized severity scores. Journal of Autism and Developmental Disorders, 44(8), 1996–2012.

Kamp-Becker, I., Albertowski, K., Becker, J., Ghahreman, M., Langmann, A., Mingebach, T., ... & Stroth, S. (2018). Diagnostic accuracy of the ADOS and ADOS-2 in clinical practice. European Child & Adolescent Psychiatry, 27(9), 1193-1207.

Karaminis, T., Gabrielatos, C., Maden-Weinberger, U., & Beattie, G. (2023). Portrayals of autism in the British press: A corpus-based study. Autism: The International Journal of Research and Practice, 27(4), 1092–1114. https://journals.sagepub.com/doi/10.1177/13623613221131752?

Lai, M. C., Kassee, C., Besney, R., Bonato, S., Hull, L., Mandy, W., ... & 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(10), 819-829.

Lord, C., Brugha, T. S., Charman, T., Cusack, J., Dumas, G., Frazier, T., ... & Veenstra-VanderWeele, J. (2020). Autism spectrum disorder. Nature Reviews Disease Primers, 6(1), 5.

Lord, C., Petkova, E., Hus, V., Gan, W., Lu, F., Martin, D. M., ... & Risi, S. (2012a). A multisite study of the clinical diagnosis of different autism spectrum disorders. Archives of General Psychiatry, 69(3), 306-313. https://jamanetwork.com/journals/jamapsychiatry/fullarticle/1107413

Lord, C., Rutter, M., & Le Couteur, A. (1994). Autism Diagnostic Interview–Revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. Journal of Autism and Developmental Disorders, 24, 659–685.

Lord, C., Rutter, M., DiLavore, P. C., Risi, S., Gotham, K., & Bishop, S. L. (2012b). Autism Diagnostic Observation Schedule (2nd ed.; ADOS-2 manual, Part 1: Modules 1–4). Western Psychological Services.

Lundqvist, L. O., & Lindner, H. (2017). Is the Autism-Spectrum Quotient a Valid Measure of Traits Associated with the Autism Spectrum? A Rasch Validation in Adults with and Without Autism Spectrum Disorders. Journal of Autism and Developmental Disorders, 47(7), 2080–2091. https://link.springer.com/article/10.1007/s10803-017-3128-y

Maddox, B. B. (2020). The accuracy of the ADOS-2 in identifying autism among adults with complex psychiatric conditions. In F. Volkmar (Ed.), Encyclopedia of autism spectrum disorders. Springer.  

McManus, S., Bebbington, P., Jenkins, R., & Brugha, T. (Eds.). (2016). Mental health and wellbeing in England: Adult Psychiatric Morbidity Survey 2014. NHS Digital. https://digital.nhs.uk/data-and-information/publications/statistical/adult-psychiatric-morbidity-survey/adult-psychiatric-morbidity-survey-survey-of-mental-health-and-wellbeing-england-2014

National Institute for Health and Care Excellence. (2012). Autism spectrum disorder in adults: Diagnosis and management (NICE Guideline CG142). https://www.nice.org.uk/guidance/cg142

NHS England. (2026). Autism diagnostic rates. https://digital.nhs.uk/data-and-information/data-tools-and-services/data-services/neurodevelopmental-data-hub/autism-diagnostic-rates

Nickel, K., Maier, S., Endres, D., Joos, A., Maier, V., Tebartz van Elst, L., & Zeeck, A. (2019). Systematic review: Overlap between eating, autism spectrum, and attention-deficit/hyperactivity disorder. Frontiers in Psychiatry, 10, Article 708. https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2019.00708/full

Ophir, Y., Rosenberg, H., Tikochinski, R., Dalyot, S., & Lipshits-Braziler, Y. (2023). Screen time and autism spectrum disorder: A systematic review and meta-analysis. JAMA Network Open, 6(12), e2346775.

Ritvo, R. A., Ritvo, E. R., Guthrie, D., Yuwiler, A., Ritvo, M. J., & Weisbender, L. (2008). A scale to assist the diagnosis of autism and Asperger’s disorder in adults (RAADS): A pilot study. Journal of Autism and Developmental Disorders, 38(2), 213-223.

Russell, G., Stapley, S., Newlove‐Delgado, T., Salmon, A., White, R., Warren, F., ... & Ford, T. (2022). Time trends in autism diagnosis over 20 years: a UK population‐based cohort study. Journal of Child Psychology and Psychiatry, 63(6), 674-682. https://acamh.onlinelibrary.wiley.com/doi/10.1111/jcpp.13505

Ruzich, E., Allison, C., Smith, P., Watson, P., Auyeung, B., Ring, H., & Baron-Cohen, S. (2015). Measuring autistic traits in the general population: A systematic review of the Autism-Spectrum Quotient (AQ). Molecular Autism, 6, Article 2. https://link.springer.com/article/10.1186/2040-2392-6-2

Santomauro, D. F., Erskine, H. E., Herrera, A. M. M., Miller, P. A., Shadid, J., Hagins, H., ... Sankararaman, S. (2025). The global epidemiology and health burden of the autism spectrum: Findings from the Global Burden of Disease Study 2021. The Lancet Psychiatry, 12(2), 111–121. https://www.thelancet.com/journals/lanpsy/article/PIIS2215-0366(24)00363-8/fulltext

Silberman, S. (2015). NeuroTribes: The legacy of autism and how to think smarter about people who think differently. Allen & Unwin: London.

Tromans, S. J., Morgan, Z. L., Brewer, B., Williams, R., Brugha, T. S., & McManus, S. (2026). Trends in self-reported autism among adults in England: Analysis of a repeated cross-sectional patient survey series of 5,999,433 adults. Journal of Autism and Developmental Disorders. Advance online publication. 

Wigham, S., Rodgers, J., Berney, T., Le Couteur, A., Ingham, B., & Parr, J. R. (2019). Psychometric properties of questionnaires and diagnostic measures for autism spectrum disorders in adults: A systematic review. Autism, 23(2), 287-305.

Williams, D., & Brick, J. M. (2018). Trends in US face-to-face household survey nonresponse and level of effort. Journal of Survey Statistics and Methodology, 6(2), 186-211.

Wing, J. K., Cooper, J. E., & Sartorius, N. (1974). Measurement and classification of psychiatric symptoms: An instruction manual for the PSE and Catego program. Cambridge University Press.

Wing, L., Leekam, S. R., Libby, S. J., Gould, J., & Larcombe, M. (2002). The Diagnostic Interview for Social and Communication Disorders: Background, inter-rater reliability and clinical use. Journal of Child Psychology and Psychiatry, 43, 307–325.

World Health Organization. (1993). The ICD-10 classification of mental and behavioural disorders. World Health Organization. https://www.who.int/publications/i/item/9241544554

World Health Organization. (2024). Clinical descriptions and diagnostic requirements for ICD-11 mental, behavioural and neurodevelopmental disorders. World Health Organization. https://www.who.int/publications/i/item/9789240077263

World Health Organization. (In press). Schedule for Clinical Assessment in Neuropsychiatry Version 3. WHO Geneva.

Yoshinaga, K., Egawa, J., Watanabe, Y., Kasahara, H., Sugimoto, A., & Someya, T. (2023). Usefulness of the Autism Spectrum Quotient (AQ) in screening for autism spectrum disorder and social communication disorder. BMC Psychiatry, 23, Article 831. https://link.springer.com/article/10.1186/s12888-023-05362-y

Zeidan, J., Fombonne, E., Scorah, J., Ibrahim, A., Durkin, M. S., Saxena, S., ... & Elsabbagh, M. (2022). Global prevalence of autism: A systematic review update. Autism Research, 15(5), 778-790. https://onlinelibrary.wiley.com/doi/10.1002/aur.2696

Please cite this chapter as:

Toomse-Smith, M., Morris, S., Gullon-Scott, F., Morgan, Z., Tromans, S., Brugha, T. (2026). Autism spectrum disorder. In Morris, S., Hill, S., Brugha, T., McManus, S. (Eds.), Adult Psychiatric Morbidity Survey: Survey of Mental Health and Wellbeing, England, 2023/4. NHS England.