Introduction
Autism spectrum conditions are characterised by difficulties in social communication, restricted and repetitive behaviours, and sensory hyper- or hypo-sensitivities (American Psychiatric Association
2013). For a diagnosis to be made, symptoms must also have been present from early childhood, and must not be explainable by the presence of other conditions such as global developmental delay. Slightly more than 1% of the UK population is autistic (Baron-Cohen et al.
2009), approximately four males are diagnosed per every one female (Fombonne
2009), and co-occurring conditions such as anxiety and depression are common (Hollocks et al.
2019). In recent times, autism has typically been conceptualised as a dimensional condition representing one end of a continuum, and so traits associated with autism can be measured in non-clinical populations (e.g. with the Autism Spectrum Quotient [AQ]; Baron-Cohen et al.
2001). Evidence for the veracity of this conceptualisation comes from observations that autistic traits are normally distributed throughout the general population (Ruzich et al.
2015), that autistic individuals typically score considerably higher than neurotypical individuals on autistic traits measures (Baron-Cohen et al.
2014; Greenberg et al.
2018; Ruzich et al.
2015), and that the scores of first degree relatives of autistic individuals are intermediate to those of autistic and neurotypical people (e.g. Wheelwright et al.
2010). As with autism diagnoses, males typically report higher levels of autistic traits compared to females (Ruzich et al.
2015), and positive correlations are observed between autistic traits and symptoms of anxiety and depression (Kanne et al.
2009; Rosbrook and Whittingham
2010). The exact nature of the association between autistic traits and depressive/anxious symptomatology remains unclear, as, although autistic people are known to report higher rates of depression and anxiety compared to neurotypical people (e.g. Griffiths et al.
2019), people with depression (Domes et al.
2016) and anxiety (Hoekstra et al.
2008) also report higher levels of autistic traits than neurotypical controls. However, it appears that the elevated rates of anxiety and depression in autistic adults are partially mediated by an increased occurrence of negative life experiences (Griffiths et al.
2019).
Similarly to autism, eating disorders are frequently found to be comorbid with anxiety (Bulik et al.
1997; Griffiths et al.
2019; Hudson et al.
2007; Martín et al.
2019) and depressive disorders (Griffiths et al.
2019; Hudson et al.
2007; Martín et al.
2019). Anorexia nervosa (AN) and bulimia nervosa (BN), but not binge-eating disorder (BED), also exhibit marked sex/gender differences in their prevalence. Although the sex ratio for lifetime prevalence of AN and BN is typically reported as being 10 females per 1 male (Raevuori et al.
2014), the extent of this difference is partially attenuated in epidemiological studies (see Murray et al.
2017). In contrast to what is observed for autism, symptoms of eating disorders are less likely to be recognised in males, and are more likely to remain undiagnosed (Murray et al.
2017). Eating disorders are increasingly conceptualised trans-diagnostically, whereby a core psychopathology (over-evaluation of eating, weight, shape and their control, and dietary restriction and restraint) underpins disordered eating across the diagnostic spectrum (Fairburn et al.
2003; Linardon
2017). Moreover, as with autism and autistic traits, symptoms of disordered eating can be detected in non-clinical groups, i.e. symptoms are distributed dimensionally not categorically in the population (Keel et al.
2007). As such, many people exhibit disordered eating symptoms which do not reach the diagnostic threshold (Favaro et al.
2003; Reba-Harreleson et al.
2009). Reflecting this understanding of disordered eating, composite symptom measures such as the Eating Attitudes Test-26 (EAT-26; Garner et al.
1982) are typically used in research with non-clinical populations to capture a broad range of characteristic behaviours and cognitions. Extant research indicates that sex/gender differences in disordered eating symptoms mirror those observed for clinical eating disorders (e.g. Bartholdy et al.
2017; McCabe and Vincent
2003; Yu et al.
2018), and, in a similar manner, disordered eating symptoms correlate positively with symptoms of anxiety and depression (e.g. Abrams et al.
1993; Diehl et al.
1998; McCabe and Vincent
2003; Porter et al.
2018).
Due to a certain degree of symptom overlap, an association between autism spectrum conditions and eating disorders has been posited for several decades (Gillberg
1983,
1985), and there has been renewed interest in this idea in recent years (Odent
2010; Oldershaw et al.
2011; Treasure
2013). Due to similarities regarding cognitive inflexibility, impaired social functioning, and restrictive and repetitive behaviours, it has even been suggested that AN “might be considered a female variant of the autistic spectrum” (Odent
2010, p. 79), and recent evidence suggests that autistic women with eating disorders experience these conditions as being closely intertwined (Brede et al.
2020). A systematic review by Huke et al. (
2013) reported a substantially increased prevalence (22.9%) of autism diagnoses within clinical AN and BN populations. Although a systematic review of more recent literature (Westwood and Tchanturia
2017) indicates that prevalence estimates may be lower in younger samples and research using parent-report measures, it is notable that a very similar proportion of women recruited from in-patient or day-patient specialist eating disorder services (14/60; 23.3%) scored above the clinical cut-off for autism on the ADOS-2 (Westwood et al.
2017). This suggests that even though the prevalence of diagnosed autism is elevated in eating disordered populations, many others who remain undiagnosed (Brede et al.
2020) could warrant a diagnostic assessment for autism. An autism diagnosis could potentially benefit such individuals by providing access to relevant services, increasing personal insight, and informing clinicians of their patients’ specific needs.
Several studies (e.g. Baron-Cohen et al.
2013; Hambrook et al.
2008; Tchanturia et al.
2013) have reported elevated levels of autistic traits in clinical eating disordered populations relative to healthy controls, with, a subsequent meta-analysis (Westwood et al.
2016) (n = 328 AN patients; n = 1890 healthy controls) revealing a large effect size (Cohen’s
d = 1.065, 95% CI 0.83, 1.23) and no evidence of publication bias. The assessment of autistic traits within eating disordered populations may also be of further importance, as scores on the AQ-10 (a shorted version of the AQ that is used in clinical settings; Allison et al.
2012) in a population of AN patients were correlated with higher anxiety and depression, and with lower ability to maintain close relationships (Tchanturia et al.
2013). Although AQ-10 scores did not correlate significantly with the level of disordered eating symptoms in this sample, Fornaro et al. (
2019) reported a significant (negative) correlation between AQ scores and BMI in a sample of n = 60 AN patients who were experiencing current depressive episodes. However, Solmi et al. (
2020) demonstrated significant correlations between autistic traits and disordered eating even after BMI had been controlled for statistically.
There has been debate as to the exact nature of the association between autistic traits and disordered eating, as increased rigidity may be expected with decreased BMI due to the effects of starvation on the brain (Hiller and Pellicano
2013). However, a recent longitudinal study (Solmi et al.
2020) demonstrated that elevated autistic social traits measured at 7 years of age predicted disordered eating symptomatology at age 14, whereas disordered eating symptoms at age 14 did not predict autistic social traits measured at age 16 once baseline values and covariates had been controlled for. Furthermore there was evidence that the association between autistic traits at 7 years of age and disordered eating at 14 years was dose-dependent, with those exhibiting more frequent disordered eating behaviours having higher levels of autistic traits than those whose disordered eating behaviours occurred less frequently. Taken together, these findings suggest that elevated autistic traits in childhood are a risk factor for the later emergence of disordered eating pathology (Solmi et al.
2020).
It is important to consider that the majority of research participants in studies of eating disorders are female (Murray et al.
2016), and that results may therefore not be generalisable to males (Murray et al.
2017). Furthermore, the prevalence of eating disorders in males may be rising (Mitchison and Mond
2015; Murray et al.
2017). Murray et al. (
2016, p. 414) suggest that the prevalence of eating disordered behaviour in men far outweighs the treated prevalence; these authors also suggest the reasons for this may be that males are particularly unlikely to seek treatment for eating disorders, because female-centric symptoms of such conditions in males are unlikely to be detected by health professionals (e.g. due to poor awareness, and/or because diagnostic instruments have typically been developed with stereotypical female presentations in mind), and because muscularity-oriented disordered eating is not currently considered a clinical eating disorder. Notably, Griffiths et al. (
2015) reported that being male and having high levels of self-stigma of seeking psychological help were both predictive of the presence of undiagnosed eating disorders; furthermore, there was a significant interaction between these predictor variables, with the association between self-stigma and undiagnosed eating disorders being stronger in males than in females. Furthermore, males have been reported to have relatively poor ‘mental health literacy’ (Mond
2014), and the symptom profile for eating disorders can differ considerably between males and females (Murray et al.
2017). Notwithstanding those men who present with female-typical symptoms, such as significant drive for thinness and weight loss, there is increasing evidence to suggest that many other males present with a pursuit of muscularity (Mitchison and Mond
2015; Murray et al.
2016). For instance, Mangweth et al. (
2001) showed that many male bodybuilders exhibit an obsessive pattern of eating and exercising similar to that of individuals with eating disorders but with the focus on gaining muscle rather than losing fat.
The above evidence suggests that more men experience eating disorders than is currently realised, and that this is to some degree contingent on ineffective recognition of symptoms more typically present in males, as well as the female-centricity of current diagnostic criteria (Mitchison and Mond
2015). However, the opposite pattern may be true of autism spectrum conditions. Biological variables such as differences in foetal sex hormone exposure (Auyeung et al.
2013; Baron-Cohen
2003; Baron-Cohen et al.
2015,
2019) may influence the sex/gender difference observed for diagnostic prevalence, and indeed sex/gender differences are known to be present for some core symptoms of autism. For example, autistic males have been reported to exhibit more restrictive and repetitive behaviours compared to autistic females (Wilson et al.
2016). However, stereotyping processes are also known to occur: autistic symptoms are less likely to be recognised in females than males (see Bargiela et al.
2016), and this effect appears to be particularly evident in autistic women with eating disorders (Brede et al.
2020).
A distinct female autism phenotype has been proposed, which may be frequently overlooked due to a bias in recognition for autistic features stereotypically associated with males. Experimental studies that manipulate sex/gender in vignettes of hypothetical children have shown that primary educators report being less likely to seek further support for females (as opposed to males) that exhibit symptoms typical of a female autistic presentation (Whitlock et al.
2020), and that laypeople are more likely to predict future atypicality in boys (as opposed to girls) who exhibit autistic features at an early age (Geelhand et al.
2019). Additionally, autistic females are thought to ‘camouflage’ their symptoms to a greater degree that autistic males (Lai et al.
2017), meaning that they may be less likely to come to clinical attention and so remain undiagnosed. Taken together, these observations suggest that although absolute sex differences appear to exist, the gaps between prevalence and detection may be magnified by gender stereotyping processes. For instance, if disordered eating symptoms are seen as a more socially accepted expression of autism-like traits in females (and indeed females may also be under greater pressure to conform socially), similar symptom profiles may result in different diagnoses in men and women.
Considering that autism may be underdiagnosed in females, that eating disorders may be underdiagnosed in males, and that clinical samples likely incur sampling bias, it is necessary to examine associations between autistic traits and disordered eating in non-clinical populations. Positive correlations have been reported between the AQ and the EAT-26 in a sample of 132 (61 males, 71 females) 11- to 14-year-old school children without recorded psychiatric diagnoses (Coombs et al.
2011), and in a sample of 80 (equal number of males and females) 18- to 25-year-old university students (Carton and Smith
2014). Mansour et al. (
2016) then reported a similar correlation in a larger sample (n = 416; 82% female) of university students, as well as evidence to suggest the association was partially mediated by emotion dysregulation (but not by negative attitudes toward emotional expression). Although Raynal et al. (
2016) did not observe a statistically significant difference in AQ-10 scores between French university students scoring above 20 on the EAT-26 (n = 101; 87.3% females, 12.7% males) and the rest of their sample (n = 378; 77.2% females, 21.8% males), the effect was in the expected direction (
d = 0.16), and a recent systematic review (Christensen et al.
2019) concluded that a positive association does exist between these variables.
Given that there are widely reported sex/gender differences in the prevalence of autism and eating disorder diagnoses, it is noteworthy that only one study (Solmi et al.
2020) has so far examined associations between autistic traits and disordered eating in males and females separately. It also remains unclear whether associations between autistic traits and disordered eating are independent of anxiety and depression. To address these gaps in the literature, we present the findings of a pre-registered study (
https://osf.io/mvx75/) that examined the potential moderating influences of sex/gender, anxiety, and depression, on the association between autistic traits (AQ) and disordered eating behaviour (EAT-26). Our main hypotheses were that AQ and EAT-26 scores would be positively correlated in both males and females, and that the association would be independent of concurrent symptoms of anxiety or depression. We assessed over-evaluation of weight and shape as a covariate in these analyses because the construct may help differentiate between obsessional/rigid approaching to eating that are characteristic of AN and BN (and possibly autism) from those characteristic of BED. Considering the apparently strong symptom overlap in women (Brede et al.
2020), we predicted that the association between autistic traits and disordered eating would be stronger in females than males.
Discussion
The main aim of the current study was to investigate associations between autistic traits and disordered eating in a general population sample of males and females. In accordance with previous findings, we observed statistically significant sex/gender differences for the AQ, EAT-26, Over-evaluation of Weight and Shape, and HADS Anxiety. Notably, however, we did not detect a significant sex/gender difference for HADS Depression, though this finding should be considered in light of the low internal consistency (α = 0.500) observed for this measure in the current study. As predicted, statistically significant (positive) correlations were detected between almost all combinations of these variables. However, the findings of greatest interest were that we observed a significant positive correlation between AQ total score and EAT-26 total score in females (rparial = 0.322, p < 0.001) and a similar trend towards significance in males (rpartial = 0.111, p = 0.087), that the correlation was significantly stronger in females than in males, and that it remained statistically significant in the combined (i.e. male + female) sample after controlling for concurrent anxiety and depression.
Previous research (Carton and Smith
2014; Coombs et al.
2011; Mansour et al.
2016) has reported statistically significant correlations between autistic traits (AQ total score) and disordered eating (EAT-26 total score) in non-clinical samples. However, other than for one recent study (Solmi et al.
2020) there has been little consideration for a possible moderating effect of sex/gender. The study by Solmi et al. (
2020) reported that the pattern of association was similar in boys and girls, though notably lacked the statistical power required to examine associations by frequency of disordered eating behaviour in males. The current study is therefore the first to show that this association is significantly stronger in females than in males. These findings are important considering that autism spectrum conditions and eating disorders both exhibit marked sex/gender differences in their diagnosis rates, that males are typically underrepresented in eating disorder research (Murray et al.
2016), and that females are underrepresented in autism research (Lai et al.
2015). In particular, these findings may be relevant when considering the propensity for symptoms to go undetected, unrecognised, and undiagnosed in autism (females > males) and eating disorders (males > females).
Although it may be that autistic traits and disordered eating really are more strongly correlated in females than in males, sex/gender differences in the phenotypic expression of eating disorders (and autism) may be relevant in explaining this observation. For instance, it has been suggested that the typical eating disorder symptom profile in males may reflect a drive toward muscularity rather than the drive toward thinness more typically observed in females (Murray et al.
2016), and that processes involved in the development and maintenance of eating disorders in autistic women may be quite different from those typically observed in non-autistic women (Brede et al.
2020). It is also relevant to note that disordered eating screening measures such as the EAT-26 often include female-centric questions that may not resonate strongly with males (Darcy et al.
2012). An obvious extension to the current research would therefore be to examine whether autistic traits are correlated with symptoms of muscularity-oriented disordered eating, and, if so, whether this correlation is conversely stronger in males than in females.
Previous studies (Carton and Smith
2014; Coombs et al.
2011; Mansour et al.
2016) have reported positive correlations between various combinations of AQ subscales (Social Skill, Attention Switching, Attention to Detail, Communication, Imagination) and EAT-26 subscales (Dieting, Bulimia and Food Preoccupation, Oral Control). This has led to some speculation about which domains may be most integral in explaining the overall association between autistic traits and disordered eating. However, our study casts doubt on there being strong domain specificity, as statistically significant positive correlations were observed between all combinations of these subscales except for AQ Imagination and EAT-26 Oral Control (
rpartial = 0.040,
p = 0.332. For the inter-subscale correlations that were statistically significant, the strength was from negligible (lowest
rparial = 0.093,
p = 0.022 for AQ Communication and EAT-26 Oral Control) to small (highest
rpartial = 0.205,
p < 0.001 for AQ Attention to Detail and EAT-26 Dieting), and none was as strong as the correlation between AQ total score and EAT-26 total score (
rpartial = 0.254,
p < 0.001). We also observed a positive correlation between AQ total score and Over-evaluation of Weight and Shape (
rpartial = 0.199,
p < 0.001); as with the correlations for the EAT-26, this effect appeared not to be domain specific, with statistically significant positive correlations of small magnitude being observed between Over-evaluation of Weight and Shape and each of the AQ subscales other than Imagination.
It has been suggested that social communication difficulties present from an early age may lead to anxiety and depression, which may then lead to the development of disordered eating (Solmi et al.
2020), and also that dietary restriction might be used by some autistic women as a maladaptive coping strategy to reduce anxiety (Brede et al.
2020). However, our study found that AQ total score remained a statistically significant (positive) predictor of disordered eating behaviour after controlling for anxiety and depression. Considering that autistic traits, disordered eating, anxiety, and depression have all be reported to correlate positively with each other, it is important to have determined that AQ scores are independently associated with disordered eating, and that this effect is not simply a statistical artefact. Essentially, whereas previous studies have established the existence of these intercorrelations, we show here that the association between autistic traits and disordered eating remains statistically significant once anxious/depressive symptomatology is held constant. These findings are therefore similar in nature to those of Mealey et al. (
2014), who established that anxiety and depression do not account for the overlap between AQ scores and schizotypal traits, and suggest that the correlation between autistic traits and disordered eating cannot be explained by there being a simple lack of specificity in the AQ (see Hoekstra et al.
2008). Awareness of the current findings may encourage practitioners to consider autistic traits in patients displaying disordered eating symptoms. As a result, treatment may be individualised to patients with elevated levels of both autistic traits and disordered eating, potentially bringing clinical benefits such as reduced recovery periods and prevention of symptom escalation. This is of particular relevance when considering the observation of Tchanturia et al. (
2016) that AN patients with low levels of autistic traits responded positively to group Cognitive Remediation Therapy (CRT), whereas those with high levels of autistic traits did not. Such findings led Westwood and Tchanturia (
2017, p. 1) to conclude that “co-morbid AN and ASD may require more intensive treatment or specifically tailored interventions”, and suggest that a better understanding of the association between these conditions is required if more effective clinical interventions are to be developed.
Strengths and Limitations
There are several strengths to our study. First, our large sample provided sufficient statistical power to reliably detect small to medium sized effects. Second, we incorporated a relatively well-balanced number of males (n = 267, 38.9%) and females (n = 419, 61.1%), which enabled us to examine sex/gender as a moderator. Third, we controlled for other relevant covariates, including anxiety and depression; this is an important consideration because, although anxiety and depression are known to be associated with both autistic traits and disordered eating, most previous studies have not included them in their statistical models. Fourth, we implemented Open Science practices by pre-registering our hypotheses and analysis plan before collecting data.
The current study also incurred some limitations. Firstly, it relied on self-report measures of autistic traits, anxiety, depression, and disordered eating. Although this facilitated collection of a relatively large dataset, it does mean that we relied on participants being able to provide accurate accounts of their current mental health status. This is an important consideration because, although some people are insightful and self-aware, others may underestimate or overestimate the severity of their current symptoms. Future research may provide more accurate insights into the nature of the association between autistic traits and disordered eating by interviewing participants’ relatives and health professionals to obtain reliability estimates. Additionally, it was noted that the majority of missing data for the current study came from participants who completed the survey via pen and paper rather than online. Although this is consistent with the observation that electronic surveys tend to result in fewer missing data than do print surveys (Boyer et al.
2002), this effect may have been exacerbated by the fact that the online study prompted participants to provide answers when relevant, whereas this was not possible in the pen and paper version. It may therefore be beneficial to use electronic as opposed to pen and paper surveys in this area of research.
Another limitation to our study is that inclusion criteria required participants to be at least 18 years of age. Although this is standard practice in much psychological research, it may be a specific problem in regard to eating disorders. This is because adolescent females represent a group at particularly high risk of experiencing eating disorders (e.g. Smink et al.
2012), yet many of those within this demographic would have been ineligible to take part in the current study. Excluding this younger age group is therefore likely to have resulted in an underestimation of the prevalence of eating disorders found in the general population. The generalisability of our findings may also have been reduced somewhat by the relatively high prevalence of Psychology students within our sample, and also by our deliberate oversampling of males from specific university courses and societies.