Introduction
Autism Spectrum Disorder (ASD) is one of the most common neurodevelopmental conditions with a prevalence of 1–1.5% of children and adults (Baird et al.
2006; Brugha et al.
2011; Christensen et al.
2016). A consistent finding from both clinical observation and empirical evidence is that more males than females are diagnosed with ASD, and current estimates range from 3:1–4.3:1 across the autism spectrum (Loomes et al.
2017). This ratio, however, varies as a function of IQ, with prevalence rates of 5.75:1 males: females in samples composed of individuals in the normative IQ range (> 70) compared to 1.9:1 in ASD associated with low IQ (≤ 70) (Baird et al.
2006; Fombonne
2009; Scott et al.
2002; Kim et al.
2011). The reason for this discrepancy in the sex ratio is unclear. While some have suggested that females may require a greater genetic load to develop ASD (Jacquemont et al.
2014), others have proposed that the male-preponderance in ASD prevalence, particularly at the intellectually able end of the spectrum, may be related to females being better at compensating for their difficulties (“female camouflage”); (Attwood
2006; Lai et al.
2011; Postorino et al.
2015; Rynkiewicz et al.
2016), potentially leading to under-recognition of females and delay in diagnosis (Lai et al.
2015). Indeed, there is evidence from population studies that girls with comparable levels of symptoms to boys are less likely to be diagnosed or are later diagnosed by community services (Russell et al.
2011; Kirkovski et al.
2013), unless they present with more substantial behavioural and/or cognitive difficulties (Dworzynski et al.
2012).
The way the core clinical symptoms of ASD—difficulties in social communication and interaction and the presence of restricted, repetitive, behaviours and interests and atypical responses to sensory input (DSM-5, American Psychiatric Association
2013)—manifest may also be different for males and females (Mandy et al.
2012; Van Wijngaarden-Cremers et al.
2014). Yet, in contrast to the strong evidence of sex differences in the prevalence of ASD, differences between the sexes in the phenotypic presentation of ASD have been found to be small in magnitude and available findings are inconsistent, both in terms of the severity of core symptoms and across age and level of functioning. While some studies have found no significant sex differences in the behavioural presentation of ASD using the ADOS (Lord et al.
2000,
2012; Ratto et al.
2017) and/or ADI-R (Rutter et al.
2003; Holtmann et al.
2007; Pilowsky et al.
1998; Andersson et al.
2013; Reinhardt et al.
2015; Harrop et al.
2015; Ratto et al.
2017), others have reported some differences using a mixed set of measures (for reviews see Lai et al.
2015; Kirkovski et al.
2013; Van Wijngaarden-Cremers et al.
2014). For example, a meta-analysis of smaller-scale studies (Van Wijngaarden-Cremers et al.
2014) and multi-site large-scale studies (Mandy et al.
2012; Szatmari et al.
2012; Frazier et al.
2014; Supekar and Menon
2015; Charman et al.
2017) demonstrated fewer restrictive and repetitive behaviours (RRB) in females than males, consistent with findings both in young children with varying cognitive abilities (Lord et al.
1982; Hartley and Sikora
2009) and intellectually able adults (Wilson et al.
2016; Lai et al.
2011). In contrast, specific sex differences in the severity of social and communication impairments have not been conclusively presented. Some studies have found girls to have more impaired social and/or communicative functioning than boys (Hartley and Sikora
2009; Carter et al.
2007), whereas others have found equivalent (Wilson et al.
2016; Mandy et al.
2012; Supekar and Menon
2015) or superior social and communication skills in females compared to males (Lai et al.
2011; Park et al.
2012). Comparisons between studies are compromised by a number of factors that potentially contribute to the discrepancy in findings.
First, females with ASD are often underrepresented due to small sample sizes that result in limited statistical power to detect small to moderate effects. Studies involving intellectually able adolescents and adults are particularly affected by this problem, and while some have addressed this issue by analysing large-scale datasets (Mandy et al.
2012; Frazier et al.
2014; Howe et al.
2015; Supekar and Menon
2015; Wilson et al.
2016; Charman et al.
2017), these studies have been limited. Second, although the ASD phenotype may present differently in males and females, current defining (DSM) criteria are still mainly based on male characteristics. This is true from both a qualitative and a quantitative point of view, because diagnostic thresholds are similar in males and females (Tsai and Beisler
1983; McLennan et al.
1993; Holtmann et al.
2007; Lai et al.
2015). This poses several problems. If current diagnostic criteria are more tuned to the male phenotype of ASD, the diagnosis of ASD in females may be missed or the condition could be misdiagnosed (Rivet and Matson
2011; Begeer et al.
2013; Dworzynski et al.
2012), even if these females present with a substantial clinical burden and would benefit from support programmes. Moreover, since sex differences in presentation may not lead to a diagnosis in females, many ASD samples potentially miss a large number of females resulting in an overrepresentation of males in ASD research even if a small group of females is included (but underrepresented) who fulfil DSM criteria, although results are thought to be applicable to both sexes (Lai et al.
2015).
Third, there is evidence that ASD symptoms may present differently across development. Some studies highlight reduced ASD symptoms with age, particularly in early childhood, but also marked heterogeneity in the trajectory of symptom expression over childhood and into early adolescence with some individuals having relatively stable high or low symptom levels across age, while others improve or become more impaired over time (Bölte and Poustka
2000; Szatmari et al.
2009,
2015; Fountain et al.
2012; Gotham et al.
2012; Lombardo et al.
2015; Bal et al.
2015). Core symptoms also often persist into adulthood, but often improve compared to adolescence (Billstedt et al.
2007; Shattuck et al.
2007; Howlin et al.
2013). Thus, comparing samples of young children (Hartley and Sikora
2009; Carter et al.
2007) to subjects across a broad age range (Pilowsky et al.
1998) may mask sex differences due to developmental changes.
Fourth, differences between males and females in the behavioural presentation of ASD may also vary with IQ, and whilst some studies have matched for IQ and age, others have not. Finally, previous studies have differed in the choice of measures used, from structured caregiver interviews (ADI-R), clinician rated observational measures (ADOS), to parent- or self-reported questionnaires, and this may have contributed to the discrepant findings (Lemler
2012; Grantham et al.
2011). The ADI-R for example probes about an individual’s current or past behaviour (ever and at 4-to-5-years—considered historically to be the ‘prototypic age’ of presentation), while the ADOS measures current symptom severity in a standardised behaviour sampling context. These instruments are relevant in our clinical and conceptual understanding of ASD symptomatology, but may yield different insights into the ASD phenotype based on their relative strengths and weaknesses in assessing symptom presentation at different developmental time-points using different informant and context-dependent assessment techniques (Charman and Gotham
2013).
Given these confounds, the pattern of sex differences in the core symptomatology of ASD remains unclear, potentially contributing to a male-bias in our understanding of ASD (for a recent special issue on this topic see Mandy and Lai
2017). One potential avenue to advance our understanding is to obtain large-scale samples which are difficult to acquire from one site alone. While some efforts are underway to actively pool clinical data from multiple sites for informative analysis (Simons Simplex Collection, Frazier et al.
2014), similar large-scale collaborative efforts have so far been largely neglected in Europe (but see Bildt et al.
2015). In response, we set up a collaboration to collect historical clinical data from ASD clinical and research institutions across Europe that are part of the EU-AIMS Clinical Network (
https://www.eu-aims.eu/clinical-network/) to examine differences across the ASD phenotype according to sex and age including larger sample sizes of females with ASD than previously examined. This circumvents the previous limited size of populations studied, narrow age ranges, level of abilities and ascertainment differences. While our primary aim was to investigate sex differences in ASD symptomatology, the size of this cross-sectional dataset and broad age distribution also afforded to analyse differences in symptomatology relating to age.
Discussion
This study investigated sex- and age-related differences in core ASD symptomatology as measured by the ADI-R and ADOS in a large and heterogeneous sample of 2684 individuals with ASD seen across 28 European clinical and research sites. Consistent with a meta-analysis of small-scale studies (Van Wijngaarden-Cremers et al.
2014) and findings from large-scale studies (Mandy et al.
2012; Szatmari et al.
2012; Frazier et al.
2014; Supekar and Menon
2015; Wilson et al.
2016; Charman et al.
2017), we found evidence of a lesser reported level of early childhood RRB on the ADI-R in females compared to males alongside comparable levels of reciprocal social interaction and communication difficulties at this age of presentation. In contrast to the present findings, some studies have also identified differences between girls and boys in early social symptoms on the ADI-R (Carter et al.
2007), but these findings are more limited and tended to report null effects when taking account of IQ (Banach et al.
2009; Lord et al.
1982).
While the overall patterns of results were maintained when non-verbal intellectual functioning was accounted for in the analyses, the significant finding of lower RRB in females relative to males dropped to a trend level after Bonferroni correcting for multiple comparisons. This makes the interesting proposition that non-verbal intellectual functioning can account and may attenuate some of the sex differences found in RRB in ASD. Alternatively, the lower significance level may also be related to a loss in statistical power due to analysing a smaller sample, which is supported by the observation that effect size estimates of sex comparisons were equivalent between the analyses. Note that regardless of whether age was accounted for in the analyses or not, the findings remained unchanged, suggesting that in this heterogeneous sample studied here, the presence/absence of sex differences in ASD severity was independent of age.
On current measures of RRB based on both caregiver interview and direct observation data, females showed as severe symptoms as males. This is at odds with some existing data demonstrating fewer current symptoms of RRB in females relative to males as measured by the ADOS (Bölte et al.
2011; Lai et al.
2011). One possible reason for differences in results may be the smaller sample size and narrower age range of the samples studied, i.e. adolescents (N = 56; Bölte et al.
2011) and adults-only (N = 83; Lai et al.
2011), compared to the much larger sample and broader age range reported in the present study from early childhood to adulthood. This may suggest that our sample composition obscured any age-dependent sex differences in RRB in adolescence and adulthood. While we did observe a significant sex by age interaction for RRB measured by the ADOS, supporting this suggestion, the results were not robust and likely the result of a small proportion of older male subjects with more severe RRB. Due to limited data points in this older age group, we were however unable to further test this hypothesis. It is important to point out that the present findings of equivalent RRB in females relative to males on the ADOS are consistent with other large-scale studies with similar age distributions (Charman et al.
2017; Frazier et al.
2014) and a recent study in adults with ASD (Wilson et al.
2016: sample N = 1244 adults with ASD; inter-quartile age range: 22–39 years). This potentially indicates that some of the previous findings of sex differences in current symptoms of RRB in adolescence and adulthood may have been sample- and/or study-specific. No sex differences relating to current social communication symptoms, as captured by the ADOS (CSS social affect) and ADI-R (social and communication domain scores), and overall ASD severity (ADOS CSS total) were observed. While this contradicts some reports of greater socio-communication difficulties on the ADOS in females (Carter et al.
2007; Hartley and Sikora
2009; Frazier et al.
2014), it is in line with others that identified no differences between the sexes (Holtmann et al.
2007; Bölte et al.
2011; Mandy et al.
2012; Reinhardt et al.
2015).
This study adds to the now growing literature that suggests that girls with ASD tend to show lesser levels of restricted interests, behaviours and stereotypes during the most ‘abnormal’ or ‘prototypic age’ of presentation, i.e. ever and 4-to-5-years, but exhibit a more similar autistic phenotype to boys in relation to social communication deficits both at younger and older ages. However, in the absence of longitudinal data in this study, conclusions about symptom trajectory or developmental changes should be considered with caution.
The current findings therefore indicate the presence of specific sex-related differences in the early developmental pattern of repetitive behaviours, routines and/or interests. What may be the factors that underlie this finding? One possibility could be etiologic protective factors, such that females have a higher liability threshold for expressing ASD symptoms compared to males, particularly for RRB (Szatmari et al.
2012). This is also consistent with behavioural genetic studies (Ronald et al.
2006; Robinson et al.
2016) highlighting the possibility for sex-and domain-specific protective factors (Constantino and Charman
2012,
2016). In the context of the skewed sex ratio in ASD towards a greater preponderance of males over females, a higher liability threshold for expressing RRB, particularly in higher-ability females with ASD, may contribute to the commonly reported widening of the sex ratio particularly at the intellectually able end of the spectrum.
Aside from a differential liability threshold, it may also be possible that higher-ability females are being under-identified as a result of displaying fewer RRB even if they present with considerable difficulties across other domains. This is in line with suggestions that clinicians are reluctant to consider a diagnosis of ASD without the presence of RRB (Mandy et al.
2012), and is reflected by the requirement for an ASD diagnosis in the DSM-5 for the presence of at least two significant indications of RRB, which is putting females at even greater risk of being unnoticed (Mandy et al.
2011). Alternatively, girls may simply exhibit ‘different’ rather than ‘fewer’ RRB than males which are therefore discounted during clinical and diagnostic assessments (Lai et al.
2015; see special issue in Autism; Mandy and Lai
2017). Clearly, future studies of the specific symptom patterns of females and how this relates to DSM-5 criteria are needed. Furthermore, early descriptions of ASD tended to be male-focussed (Kanner
1943) and diagnostic instruments including the ADI-R and ADOS were predominantly developed using male samples, leading potentially to a male-biased understanding of ASD and concomitant sex bias in the construct and item-structure of the instruments themselves. This may suggest that future revisions of these instruments require additional items to be included that are more characteristic of the female ASD phenotype. At least for the ADI-R, there is some evidence to suggest equivalent scale and item structure of the ASD phenotype in males and females (Duku et al.
2013; Frazier and Hardan
2017), but such evidence is missing for the ADOS. A future goal of research should therefore be continued exploration of the psychometric properties of these instruments (including establishing measurement equivalence across sexes) to evaluate the requirement for sex-specific norms (Constantino and Charman
2016; Lai et al.
2015). Future studies will also benefit from investigating sex differences using instruments that might be more sensitive to potential sex differences in presentation of ASD characteristics also outside of the clinical arena, such as the SRS-2 (Constantino
2012), a parent, teacher, spouse, and/or self-report questionnaire measure of autistic—like traits (Frazier et al.
2014; Howe et al.
2015; Charman et al.
2017; Ratto et al.
2017), compared to the ‘gold-standard’ diagnostic instruments the ADI-R and ADOS used in the current study.
Another possibility for the current results is that rater reports may have influenced the findings. Mothers are typically the primary source of information during diagnostic assessments and sex differences reported on the ADI-R may be a function of parents reporting symptoms differently for girls and boys. In the current study however, we were unable to further assess these possibilities. Lastly, the current results may also potentially reflect sex differences in RRB in early typical development. However, while some studies have found boys to score higher than girls on ratings of repetitive behaviours and preoccupations with restricted patterns of interest, but not repetitive movements, sensory interest, or rigidity (Leekam et al.
2007), others have not demonstrated sex differences in RRB in early development (Evans et al.
1997; Øien et al.
2017).
Age-related analyses revealed lower current social and communication symptoms with age as measured by the ADI-R, both with and without covarying for NVIQ, with older subjects reporting lower symptom scores than younger subjects. Since the majority of participants fell within the 2–25 years’ age range, beyond which data was more limited, the significant differences in symptom scores as a function of age largely reflected differences across this particular age range rather than the entire sample. ADOS CSS total and CSS social affect displayed a similar albeit attenuated effect of a negative relationship between symptom scores and age, which however disappeared when non-verbal intellectual functioning was accounted for in the analyses. These results broadly support a range of studies showing reduced ASD symptoms with increasing age, including those studies that tracked samples longitudinally since childhood (Billstedt et al.
2007; Howlin et al.
2013; Shattuck et al.
2007). Larger cross-sectional samples that have also reported differences in symptomatology with age are rare, but those that did, did not find significant age differences on the ADOS when IQ was included in the model (e.g. N = 325, Mandy et al.
2012; N = 437; Charman et al.
2017). Given the cross-sectional nature of the data, it is not clear if the age-related differences observed reflect true effects or are due to sampling differences between datasets that recruited participants across different ages.
Limitations
Although the total sample size of the current study was large, the sample consisted of individual datasets pooled across many different sites that were not fully matched for assessment methodologies, diagnostic procedures and ascertainment strategies. Also, samples were derived across different research programmes with different purposes (e.g. early screening studies, intervention programs, high-risk sibling studies, genetic and imaging studies), and differed in respect to the distribution and range of ASD symptom severity, age and intellectual functioning. However, unfortunately, the individual sample sizes for each dataset were too small to allow for any additional meaningful comparisons within individual datasets.
It is also important to acknowledge that for data relating to the ADOS, participants were not equally distributed across the different modules, with the majority of subjects completing Module 1 designed for individuals who are preverbal or who use single words to communicate. This somewhat limits the conclusions drawn in relation to age-related trends in the ADOS data.