Background
Eating disorders (EDs) are characterised by markedly abnormal attitudes to body weight and food intake, which result in disturbed patterns of eating and behaviour. They include anorexia nervosa (hereafter ‘anorexia’), bulimia nervosa (‘bulimia’) and eating disorders not otherwise specified (EDNOS) [
1]. Anorexia is diagnosed when a person becomes substantially underweight due to restricted eating, arising from a fear of putting on weight and a distorted body image [
1]. Bulimia is defined as recurrent episodes of binge eating followed by inappropriate compensatory behaviour (for example, vomiting, laxative use, excessive exercise) in order to prevent weight gain [
1]. EDNOS describes eating difficulties that are clinically severe, but which do not fulfil criteria for a specified ED such as anorexia or bulimia. Autism spectrum disorder (ASD) is a neurodevelopmental condition, characterised by pervasive difficulties with social reciprocity, social communication and flexibility [
1].
At first glance, EDs and ASD would appear to have little in common. ASD is a disorder of social function and flexibility, which manifests in the first year of life. By contrast, bulimia and anorexia are concerned with abnormal eating behaviour, with typical onset around adolescence and early adulthood [
2]. Males are at greater risk of ASD than females, whereas EDs shows the converse gender ratio, affecting 10 females for every male [
3,
4]. While one-third of people with ASD have an intellectual disability, EDs are not related to intellectual impairment; and anorexia may actually be associated with above average intelligence [
5,
6]. ASD is a lifelong condition; bulimia and anorexia fluctuate across the lifespan [
7].
Despite this, there is currently interest among clinicians and researchers in the overlap between ED and ASD, with a specific focus on the hypothesis that ASD places women at high risk of developing anorexia (for example, [
8]). This idea stems from Christopher Gillberg [
9] and his collaborators’ [
10,
11] observation that many women with anorexia show inflexibility and impaired social function, and that this may reflect the presence of an underlying ASD.
Gillberg and colleagues have published evidence to support their proposal that ASD and anorexia are associated, based on findings from two samples. They conducted a long-running study of a cohort of people (94% female) with adolescent-onset anorexia in Gothenburg, Sweden, estimating ASD prevalence in this group to be between 8% and 37% (see review by Huke and colleagues [
12] for a synthesis of this work). Subsequently, the Gothenburg researchers reproduced their Swedish findings in the UK, reporting that seven of 30 women (23%) attending specialist ED clinics in London met criteria for ASD [
13]. In addition to replicating previous findings in a new sample, this study is important because it extended the observation of high prevalence of ASD to an ED sample that included women with bulimia as well as anorexia. It should be recognised that the rates of ASD reported by the Gothenburg team, found in almost entirely female samples, are strikingly high, as in the general population of females ASD prevalence is estimated to be 0.3% [
14].
In support of these diagnostic findings are studies of sub-threshold ASD symptomatology (‘autistic traits’) and of autistic cognition in anorexia. Investigations of autistic traits, measured by self-report using the Autism Quotient (AQ) [
15], have shown elevated levels of autistic symptomatology in females with anorexia compared to controls [
16‐
18]. Furthermore, at the group level, people with anorexia tend to show a cognitive profile characterised by poor theory of mind [
19], cognitive inflexibility [
20] and detail focused processing [
21]. This mirrors the profile of cognitive difficulties found in groups of people with ASD (for example, [
22]).
Nevertheless, some researchers have dismissed outright the idea that ASD is highly prevalent among people with ED [
23]. Their scepticism arises from some substantial methodological limitations of the ASD-in-ED literature. One criticism is that all the ASD diagnostic findings in adults with ED come from one group of researchers, and that five of their six relevant studies are based on the same Swedish community sample [
12].
A more profound challenge to the argument that many women with ED have ASD is that their social difficulties and inflexibility may not be truly autistic in origin. People with ED experience high rates of obsessive compulsive disorder (OCD), anxiety, depression and starvation, all of which can give rise to social impairment and rigidity that could be mistaken for symptoms of autism. The high rates of ASD reported in anorexia may reflect a problem of construct validity, with social and flexibility difficulties being mislabelled as autistic symptoms, rather than a true overlap between the two disorders [
23].
One strategy for teasing apart autistic and non-autistic symptomatology is to use gold-standard, well validated assessment measures, designed to implement DSM/ICD accounts of ASD [
24]. Such measures have not so far been used to assess adults with ED. In the initial Gothenburg anorexia studies, information was gathered in a general clinical interview, and this was then used to estimate ASD diagnosis [
10,
25]. In later Gothenburg studies [
11,
26,
27] general clinical interviews were supplemented by a structured Asperger’s assessment, the Asperger’s Syndrome Diagnostic Interview (ASDI) [
28]. While the ASDI has much promise as a measure of autistic symptoms in high-functioning individuals, ASD prevalence rates based on its use are difficult to interpret for the following reasons. First, as its name suggests, the ASDI was designed specifically for the assessment of Asperger’s syndrome, and so it may not be suited to assessing the full range of presentations encompassed by the ASD diagnostic category. Second, it implements ‘Gillberg and Gillberg’ criteria for Asperger’s [
29], which do not overlap fully with the almost universally accepted DSM and ICD accounts of ASD. Third, only preliminary information is currently available on the validity of the ASDI with respect to independent clinician diagnosis [
28]. Fourth, in the ASD-in-ED studies, the ASDI appears to have been administered using a mixture of parent report, self-report and clinician observation, whereas it was designed and validated as an informant report interview [
28]. A further complication when interpreting prevalence rates of ASD in anorexia is that multiple different definitions of ASD have been used throughout the Gothenburg studies, based on DSM-III, DSM-III-R, DSM-IV, ICD-10 and ‘Gillberg and Gillberg’ criteria [
12]. These methodological factors may partly account for the wide variation of published ASD prevalence estimates in the Gothenburg sample (8% to 37%) [
12].
The failure to use non-standardised assessments designed to implement a universally recognised (that is, DSM and ICD) definition of ASD may have led to over-estimation of the rates of ASD in ED [
23]. The one study that has used a standardised, DSM-based ASD assessment, the Dimensional, Developmental and Diagnostic Interview - short version (3Di-sv) [
30,
31], in young people with a restrictive ED did not find any evidence for high rates of ASD compared to controls [
32]. This could indicate that previous findings of Gillberg and colleagues in adults with ED are an artefact arising from the use of non-standardised assessment of ASD; or it could have instead arisen from the nature of the participants studied, who were children and adolescents with ‘early onset ED’ [
32], not adults who all met criteria for anorexia, as in the Gothenburg studies.
Thus, despite the influential work of the Gothenburg anorexia researchers (for example, [
25,
10,
13]), it is currently unclear whether the social difficulties and inflexibility observed in adolescents and adults with EDs reflect underlying autistic difficulties; or whether they arise instead for other reasons, such as the effects of starvation, anxiety, low mood and OCD. A crucial step towards resolving this uncertainty will be the use of well-validated and standardised measures of ASD in samples of adults with ED. While parent report is a cornerstone of ASD assessment in children, in clinical practice with adults it is often difficult to attain, especially for those whose psychopathology may both reflect and cause family discord [
33]. Thus we sought to pilot the use of a direct observational measure of ASD symptoms and diagnosis, called the Autism Diagnostic Observation Schedule (ADOS) [
34].
The ADOS is the gold-standard structured observation tool for diagnosing ASD [
24]. It is widely used in clinical and research practice, and can be administered with adults and adolescents as well as children. Despite this, we know of no reports of its use in adults with ED. We used the ADOS to conduct a series of clinical assessments of women with ED (anorexia, bulimia or EDNOS) whose social and/or flexibility difficulties had lead their clinical care team to suspect they had ASD. While the ADOS has shown excellent sensitivity and specificity in distinguishing adults with ASD from clinical controls, in common with all other assessments of autistic symptoms it was validated in a largely male sample [
34]. Anecdotal reports and some indirect empirical evidence suggest that it may lack sensitivity for some symptoms of ASD as they present in high-functioning women [
35]. As most people with ED are adult females with average or above IQ, we argue that the value of the ADOS as an assessment in this population cannot be assumed, making a pilot study necessary.
By offering in-depth qualitative and quantitative information about the use of this standardised direct observational assessment in clinical practice, we aimed to contribute to debates about whether the social and flexibility difficulties seen in some people with anorexia, and with ED more generally, are autistic in origin; or whether they are non-autistic difficulties that only superficially resemble ASD. By writing this case series, we also wished to present information about the feasibility of the ADOS in this population, to inform decisions about its future use in research and clinical practice with people with severe ED.
Discussion
Among people with EDs, social difficulties [
41,
42] and inflexibility [
43] are common. It is currently unclear to what extent these are manifestations of an underlying ASD; or whether they are instead epiphenomena of ED which only superficially resemble autistic symptoms [
23,
8]. A step towards resolving this clinically and theoretically important uncertainty will involve the use of gold-standard ASD assessment tools in adults with an ED. To this end, we report a case series of Autism Diagnostic Observation Schedule (ADOS) [
34] assessments, carried out with 10 women who had an ED, as well as social and flexibility difficulties that had led their clinical care teams to suspect they have ASD. This is, to our knowledge, the first report in the scientific literature of the ADOS’s use in ED, and we aim to provide information about its feasibility in this population, as well as information about ASD in our sample.
Overall, our findings are supportive of the idea that there is a genuine overlap between ASD and ED [
9,
10,
13]. Among the 10 women assessed, five met criteria for an autism spectrum disorder on the ADOS Module 4 algorithm. This reflects the fact that they demonstrated a range of social and communication difficulties that are consistent with the autistic phenotype, including unusual eye contact, limited empathy and social insight, abnormal use of gestures to compliment verbal communication and atypical intonation.
All five women scoring for ASD on the ADOS had anorexia and a body mass index below 17. One interpretation of this is that their high scores on the ADOS do not signify genuine autistic symptoms, but instead reflect the severity of their ED, including the effects of starvation. However, this possibility is countered by reports from each of these five women that their autistic difficulties had been present in childhood, prior to the onset of their ED. This is consistent with the fact that ASD is a developmental disorder that appears early in life [
44], and counters the idea that their social and flexibility difficulties are merely a consequence of their eating and emotional difficulties.
Nevertheless, more work is required to test adequately the hypothesis that ASD is a causal risk factor for ED. To this end, we suggest several research strategies. First, prospective studies are needed to investigate whether in some cases ED is preceded by ASD or autistic traits. Second, research in this area should go beyond examining symptom overlap, to consider shared underlying mechanisms of ASD and ED. The current interest in endophenotypes for mental disorders, exemplified by the NIMH’s Research Domain Criteria (RDoC) strategy [
45], can inform investigations into whether ASD and ED share common cognitive and neurobiological underpinnings. Third, family studies investigating rates of ASD in the pedigrees of people with ED, twin studies and genome-wide complex trait analysis could all be used to estimate the amount of shared genetic risk for ASD and ED.
We identified two women (P6 and P7) who did not score in the ASD range on the ADOS, but may actually have an autistic disorder. We suspect they have ASD for a number of reasons. First, they reported that in childhood they had various neurodevelopmental difficulties associated with ASD, including problems with sensory processing, emotion regulation, language and attention [
46‐
48]. Second, both were referred in childhood to psychologists for neurodevelopmental difficulties, and one actually received an ASD diagnosis at this time. Third, they described histories of childhood social difficulties, leading to social isolation and bullying which are consistent with having ASD [
49]. Fourth, they had family members with substantial neurodevelopmental difficulties, including Asperger’s syndrome. Family members of people with ASD are at substantially increased risk of ASD and other neurodevelopmental problems [
50]. Fifth, when we contacted the mothers of these women to learn more about their development, they reported a range of childhood difficulties consistent with ASD, including non-social aspects of ASD such as inflexibility and sensory abnormalities. P7 scored above the 3Di-sv thresholds for an ASD diagnosis; and P6 scored below it, but nevertheless showed elevated parent-reported levels of autistic traits.
The assessment of ASD in females with normal-range IQ is notoriously difficult, as there is a female autistic phenotype, characterised by subtler difficulties compared to equivalent males [
3,
51,
52]. This may help us understand why P6 and P7 did not score in the autistic range on the ADOS. Studies of ASD in women have suggested that compared to men, women tend to score lower on the ADOS Module 4 [
52]. This likely reflects a key element of the female ASD phenotype, which is a capacity for camouflaging autistic characteristics in social interactions [
35]. In this context, camouflaging denotes an effortful, conscious masking of underlying autistic difficulties, using imitation, reasoning and symptom suppression [
52]. It is notable that both P6 and P7 gave examples of how they have deliberately camouflaged their autistic difficulties. For example, P7 spoke about suppressing the urge to rock and flap, and P6 described how she consciously learnt how to use eye contact and gesture by imitating other girls. These findings highlight the challenges of assessing ASD in women with EDs, showing that while the ADOS is likely a useful tool in this population, as in any ASD assessment, its algorithm should never be used as the only basis on which to make a diagnostic decision.
Our findings must be considered in the light of several limitations of the current study. While the ADOS is well-designed to distinguish between autistic and non-autistic psychopathology (34), it is nevertheless possible that we have overestimated the levels of autistic symptoms in our sample, due to the presence of starvation, anxiety, depression, attention deficit/hyperactivity disorder and other difficulties that co-occur with ED at above-chance levels. One way to mitigate this problem in future would be to include a control group of women matched for ED severity and co-occurring conditions, but who do not have marked social and flexibility difficulties. Another strategy for increasing the validity of our findings would be to conduct a multi-modal assessment for all people in the study, including a standardised informant-report developmental history, and to use the resultant information to derive a clinician consensus diagnosis based on DSM-5 criteria. Any study seeking to estimate the true prevalence of ASD in ED should take this approach.
Conclusions
The findings we present come from a pilot case series, and must be considered preliminary. Nevertheless we believe they justify the following recommendations. First, we found some evidence for inadequate assessment of ASD in females, and action is required to redress this. In our study of 10 women with an ED, five scored up for ASD on the ADOS and a further two did not reach ADOS thresholds, but we nevertheless suspect that they have ASD. But of these seven women, only one had a prior autistic diagnosis. All of them reported significant social impairments in childhood before the onset of their ED, often accompanied by peer victimisation; most had histories of serious neurodevelopmental difficulties, such as dyslexia and epilepsy; and longstanding difficulties with sensory processing and flexibility were commonly reported. Several had been assessed as children by mental health professionals. And yet their ASD had been missed in childhood, and as a consequence they did not receive appropriate support and understanding at this time. These specific experiences fit with a broader picture of females missing out on ASD diagnoses and services. Females with ASD are under-represented in ASD clinics [
53]. When they are identified, this tends to be later than for males [
54] and they require more severe symptomatology than males to be meet ASD diagnostic criteria [
55]. Urgent efforts are required to elucidate the characteristics of the female autistic phenotype and to use this knowledge to reduce the current diagnostic bias against girls and women with ASD.
A second recommendation is that the ADOS is a potentially feasible and valid tool for assessing ASD in women with an ED. We found it could be administered according to protocol even with people suffering from acute and severe anorexia. We have presented qualitative evidence that the ADOS provided valuable information for diagnosing ASD; but also that scores on its diagnostic algorithm are not in any sense definitive. Our experiences suggest that, wherever possible, observations made during the ADOS should be supplemented by reports from informants, including other clinical staff and family members, and this information should be used to inform a consensus diagnosis reached between clinicians [
39].
A third recommendation is that research is required to provide clinically useful information about people who have ED and ASD. It will be important to discover the true prevalence of ASD among people with anorexia, if rates of ASD differ according to ED diagnosis, and whether individuals with ED and ASD have a distinct aetiology, treatment needs and prognosis compared to people with ED who do not have ASD. We think that further research in this direction will help us to develop more precisely targeted interventions; and can also provide information on the female phenotype of ASD.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
KT and WM designed the study together. KT obtained approval for the study, conducted recruitment and WM performed the ADOS assessments. Synthesis of information collected and writing the paper was done jointly by WM and KT. Both authors read and approved the final manuscript.