Mitigating Sex-Related Biases to Elucidate the Autism Phenotype

Abstract

Autism spectrum disorder (ASD) is commonly considered a male-dominant condition, with epidemiological estimates finding that approximately 3 to 4 males are diagnosed for every female. An overwhelming majority of the studies used to establish this sex ratio were conducted with participants ascertained based on having first met clinical criteria, which may obscure qualitative differences between males and females and omits females who do not meet male-biased criteria. Our recently published data, which used a prospectively identified sample of children at high familial likelihood for ASD, corrected for sex-based measurement bias, and used data-driven groupings of behavior over time, were well suited to address this issue and suggest that the male:female ratio of autism-related concerns is closer to 1:1. In this review, we propose that research is needed that characterizes the autism phenotype, or behavioral expression of underlying genetic variation in autism-relevant traits. We describe shifts in sample ascertainment, methodological rigor, and the scope of traits examined that may help elucidate the autism phenotype. A research program that focuses on delineating the genetically related, biological determinants of clinical disorders has the potential to reveal the full expression of underlying genetic liability for ASD and improve early identification of ASD-related concerns in females.

Longstanding conventional wisdom suggests a male:female sex ratio of 4:1 in autism spectrum disorder (ASD). This empirical finding has anchored considerable research concerning sex-based variability in behavioral/cognitive characteristics, as well as preclinical work on explanatory mechanisms that may account for the sex differences in ASD. However, most studies were conducted with participants ascertained based on having first met clinical criteria for ASD, and most studies have included primarily male participants. This has created a bias toward male presentations of ASD, and the field is perpetually stuck in a cycle of reifying a heterogeneous diagnostic category that does not map onto biological differences (1,2). In Burrows et al. (3), we illustrated a novel approach that addressed several biases in previous research. We followed a sample of younger siblings of autistic children who were prospectively recruited prior to the emergence of defining features of ASD, corrected for sex-related measurement bias, and considered sex differences broadly, regardless of level of ASD concern. In doing so, we identified an equivalent number of males and females with elevated ASD concerns, although concerns were generally milder in females. Our 1:1 sex ratio in autism concerns upended decades of research on sex differences in ASD. Sex-related biases exist in current ASD identification and research, and there is likely a greater proportion of females who exhibit autism-related concerns than epidemiological estimates would lead us to believe. These innovations can be expanded to better understand the autism phenotype, or the behavioral traits that express heritable ASD risk. Here, we call for shifts in sample ascertainment and composition, consideration of measurement bias by sex, and investigation of traits that extend beyond core DSM-defined symptoms of ASD. Coupled with epistemic curiosity, this approach has the potential to define valid phenotypes associated with genetic liability for ASD to more closely map onto underlying biology, improve rates of early detection of ASD, and develop treatments that address the range of challenges faced by females with ASD.

SEX BIASES IN ASD CRITERIA AND RATES

Sex Differences in ASD and Typical Development

When studying sex differences in ASD, the field has consistently focused on core ASD symptom domains of social communication and restricted and repetitive behaviors (RRBs), as well as the presence of related clinical impairment (e.g., cognitive level, language abilities) (4–6). Within these core domains, autistic females have been reported to show equivalent social communication concerns and slightly less severe RRBs, particularly in older childhood and adolescence (7,8). By contrast, when asked, clinicians report perceiving clear sex differences in autism presentation, particularly at later ages (9).

Many traits that are considered core to the clinical presentation of ASD differ by sex in the general population. Thus, sex differences in autistic samples may simply reflect sex differences broadly (10,11). Females show advantages in social motivation (12–14), social reciprocity (15), emotional expression and understanding (16,17), and joint attention (18,19). There has been some argument over whether these sex differences are innate or socialized (20,21), although sex differences emerge earlier in development than would be suggested from a socialization approach (14,22). Females also consistently outperform males in language tasks across development, with consistent but small differences in the timing of first words and vocabulary size (23). In adolescence and adulthood, the skills that are important for friendships also differ between males and females, highlighting the lasting nature of sex differences in social interactions (24). Typically developing boys also show more significant intense interests in early development (25,26) and repetitive motor behaviors in early childhood (27).

When examining autistic traits in the typically developing population, females demonstrate lower levels of social impairments, particularly during adolescence. Typically developing females also showed lower levels of RRBs (10). It is critical to compare autistic females with typically developing females in addition to comparing autistic females with autistic males.

ASD Prevalence Reveals Ascertainment Bias

From its conception, ASD has been considered a male-dominant neurodevelopmental disorder. In Kanner’s seminal description of autism in 1943, 8 of the 11 original cases were male (28). While Kanner did not specifically detail sex differences in symptoms, his descriptions belied sex-specific behavioral expectations, including describing Barbara K (Case 5) as having “no competitive spirit, no desire to please her teacher.” It appears that this comment is meant to highlight that Barbara was not following the social stereotype that females are other-oriented and aim to please those in authority (20,29,30). While this may indicate some consideration of a sex-related difference in what autism could look like, this description is drowned out due to a minority of cases being female.

Since Kanner’s early descriptions, the field of ASD has expanded and evolved (31), although it has generally followed the cycle of first studying individuals who have been identified as autistic and then subsequently conducting studies and updating the diagnostic criteria and instruments based on findings from these individuals. Most ASD diagnostic tools were developed and calibrated using predominantly (.75%) male samples (32,33), thereby imposing epistemic blinders of the clinical diagnostic context onto the autism research context (1). While diagnostic measures work well to differentiate ASD from other conditions in both males and females, they may be less sensitive in milder cases or individuals who are currently missed in the current clinic referral pipelines for ASD.

Clinical diagnoses are linked to impairment. Behaviors that are either shared among disorders or that are not associated with impairment are typically not incorporated into the categorical clinical definition. Similarly, behaviors that have historically been viewed as outside the realm of psychiatry (e.g., motor difficulties) are not candidates for defining clinical diagnostic criteria. These biases arising from a clinical nosological framework have skewed our perception of ASD.

Epidemiological studies of ASD prevalence have found a sex ratio of around 3 to 4 males for every female (34–38). Variations within the sex ratio can reveal where biases exist in our current system. Early research identified that the sex ratio varied based on a child’s cognitive ability level (IQ). From the 1980s to early 2000s, the sex ratio was estimated at 9 males:1 female in children with average to above-average cognitive abilities but closer to 1.6:1 in children with intellectual disability (39–41). While there may be sex differences in the relationship between autistic traits and cognitive ability (42), this finding may reflect that clinicians were missing autistic females with average and above-average IQs. Thus, the clinical perception of a symptom profile for children with average and above-average IQs may be skewed toward a male-dominant presentation. In support of this hypothesis, sex differences in ASD for children with intellectual disability continue to persist but have decreased over time (36,43). Furthermore, ASD diagnoses are increasing more rapidly in females (44), resulting in a decreasing sex ratio in recent years (36,38).

Sample ascertainment strategy is one of the strongest factors that predicts the presumed sex ratio of autism (45,46). The estimated global ASD sex ratio is 4.2:1 (35), with variations based on health care practices and differences in ASD identification practices (37). Community-referred samples and epidemiological studies include children who have been identified within the current ASD referral system and found to meet full criteria for ASD. Thus, potential sex-related biases in screening processes (47,48) or access to evaluations (49) would result in fewer females being included. For example, studies that used clinically ascertained samples and required participants to meet clinical cutoffs on ASD diagnostic instruments resulted in a higher rate of exclusion of females, with study sex ratios closer to 7:1 (46).

Population-based studies may also help explain the source of sex differences in who is missed in the current ascertainment system. A twin study that compared autistic individuals with individuals with high autistic traits but without a diagnosis showed that females needed to demonstrate additional problems to be identified as autistic (50).

By contrast, studies that prospectively monitor for ASD or that follow children who are at higher likelihood of developing ASD are less prone to ascertainment bias and tend to find lower sex ratios (45,51), closer to 2:1 (51–53). Infant siblings of autistic children are at higher familial likelihood (HL) of developing ASD. HL studies were initially used to investigate the early progression of ASD, as recurrence rates are greater than population-based epidemiological rates of ASD. Behavioral features of ASD emerge during the latter part of the first year, and a diagnostic profile typically consolidates between 18 and 36 months of age (54–56). Thus, HL studies that recruit children in infancy (up to 6 months) are recruiting children who are not showing any behavioral signs of ASD, which minimizes ascertainment biases. Sex ratios have been declining in HL samples over time [2.8:1 in 2011 vs. 1.98:1 in 2024 (49,50)], highlighting that clinicians may be improving in their ability to identify ASD very early in development in females. The removal of language delay as a symptom of ASD in the shift from DSM-IV to DSM-5 might have also reduced bias resulting from sex differences in language. Prospective universal screening studies have also found lower sex ratios of approximately 2.8:1, highlighting that actively seeking to identify autism rather than waiting for individuals to be diagnosed also mitigates sex-related biases (57).

Epidemiological and case-control studies cannot capture the autistic individuals who are missed in current referral systems or those who do not fit clearly within DSM-based criteria for ASD. In both cases, these effects may be greater in females. HL samples can address some of these limitations. Autistic HL participants tend to have lower severity of autism symptoms and higher developmental functioning than either community-referred autistic samples or samples ascertained through universal screening (51,58). This pattern is unsurprising, given that screened samples must exceed a behavioral detection threshold, whereas HL participants do not. The imposition of a screening measure gatekeeping access to identification of concern imposes an inherent bias and results in a difference in sample composition. The difference in autism severity and developmental level implies that the screening instruments skew toward identifying children with ASD with more significant developmental or language impairments.

Beyond Autism Diagnostic Criteria

Few sex differences emerge in autistic samples when broad measures of core autism traits are examined (7,8). However, in verbally fluent individuals, more subtle measures of social communication have shown that autistic females demonstrate better conversational skills (59) and use more social language (60,61), often despite comparable levels of autistic traits on the Autism Diagnostic Observation Schedule. Similar nuanced sex differences exist in levels of RRBs. Many RRB instruments focus on repetitive play, areas of interest, and rituals and routines that are related to male interests (e.g., lining up cars, strong interest in train schedules). Clinicians and autistic females have noted that repetitive behaviors or circumscribed interests with books or dolls may be undetected or misinterpreted as pretend play (62). Similarly, investigations into the types of interests that HL siblings show early in development have found that females showed different types of interests, although for autistic HL individuals, there was no sex difference in the intensity or peculiarity of interests (25). Thus, our measures may need to be revisited with a lens toward sex differences and examples of RRBs that may be more prevalent in females.

Sex differences in ASD are even more pronounced in domains that lie outside of core autism symptoms. Autistic females, relative to autistic males, have higher functional impairment, including challenges with executive function (63), emotion dysregulation (64), mental health problems (65–67), and poorer adaptive adult outcomes (68). There may be differences by sex in the phenotypic aggregation of these traits, particularly if a clinical lens is removed and these co-occurring conditions could be viewed as behavioral expressions of underlying genetic or biological differences (69). Currently, it could also be that using the same male-biased threshold for diagnosing ASD in males and females results in higher impairment for females who receive a diagnosis. This may be more pronounced in samples with average to above-average cognitive abilities, where the ascertainment bias has a more significant impact.

MITIGATING SEX-RELATED BIASES

In Burrows et al. (3), we investigated sex differences in ASD in a manner that addresses several of these challenges. First, we used a prospectively ascertained sample of HL children who were recruited in infancy, prior to the onset of any clinical concerns. In our measurement of social communication and RRB, we corrected for sex-related measurement bias to ensure that we were comparably measuring symptoms for males and females. To do so, we tested for differential item functioning in the underlying latent constructs of social communication and RRBs. Differential item functioning tests whether males and females show different mean levels of a specific trait (intercept level) or different relationships between indicators and the latent construct, given similar underlying levels of the latent construct. Accounting for measurement bias will differentially weight items that index the underlying latent construct by sex to optimize how the latent construct is being measured for both males and females. Trajectories of change were then examined from 6 to 60 months. In doing so, we identified a group of children with consistently high social communication concerns. Our high-concern group had a 1.06:1 sex ratio, demonstrating that females are equally likely to show elevated social communication traits as males when 1) ascertainment biases are reduced, 2) measurement bias is taken into account, and 3) behavior is considered over a wide developmental span. Sex differences emerged early in development, with significantly higher concerns for males regardless of concern group starting at 6 months of age.

Ascertainment Bias

Diagnostic criteria for behavioral syndromes, as defined in DSM-5 and its predecessors, do not necessarily reflect variations that map onto underlying biology or reflect natural kinds, particularly in the case of autism (2,70). The study design offers an important opportunity to mitigate biases and conduct more representative research. Astle et al. (70) describe a continuum of recruitment approaches leading from restrictive (case-control) to more population-representative research. Family designs such as infant sibling studies represent recruitment based on a functional definition (having an older autistic sibling) to mitigate ascertainment biases. While several large-scale and population-representative case-control studies exist and are available for public use, few large-scale family studies exist. The Baby Siblings Research Consortium provides an example of a group of researchers who have pooled multiple infant sibling samples and is expanding its definition of autism liability to include other groups at higher likelihood for ASD. These types of samples offer an opportunity to better understand the autism phenotype. However, HL studies may be biased in that the autistic probands were diagnosed using potentially male-biased clinical criteria, participants were often excluded if there was a known genetic disorder in the proband, and some families with an autistic child may not have other children (71) or may experience barriers to participating in research (72,73). Another approach could include recruiting based on parental or family levels of ASD traits or risk for psychopathology generally (74).

Recent innovations in the ability to sequence DNA at scale for low cost, as well as in our understanding of genetic underpinnings of behavior, have also paved the way for new recruitment approaches. Studies that conduct large-scale neonatal genetic screening and have access to later outcomes as documented in a medical record could pave the way for genetically delineated sampling. Some large-scale studies are underway to help identify potential genetic differences that could be identified in newborn screening and used to ascertain a sample to follow intensively (75).

Another approach to mitigating ASD-related ascertainment biases involves using large-scale population-representative research to uncover biologically meaningful groups. For example, Doyle et al. (76) collected parent-reported measures of behaviors common to ASD in a large community-based sample of toddlers and implemented factor mixture modeling to identify phenotypic risk profiles. Groups showed meaningful differences in internalizing and externalizing problems and dysregulation at a later time point. Similar approaches have also been conducted within clinically ascertained groups (77,78), and studies have found that behaviorally defined groups showed differential brain connectivity profiles and did not map onto DSM-defined conditions. Notably, in Kushki et al. (77), the clinically ascertained sample did not have a sufficient number of females to test for sex-related differences in groups, highlighting additional challenges to a clinical approach to sample ascertainment.

Measurement Bias

Establishing equivalence of measurement between males and females is a necessary step to examining sex differences in ASD (79,80). Most studies examine mean-level differences without evaluating that the structure of the latent factor is equivalent. If sources of bias are not considered, there may be systematic bias in comparison of males and females (81). In Burrows et al. (3), we found that initiating joint attention was a less strong indicator of social communication for females, suggesting that joint attention did not covary with other social communication features as reliably for females. This highlights that there may be slight sex differences in the underlying structure of ASD symptoms that need to be accounted for when studying overall levels of traits in ASD. For all traits of interest, it is critical to consider dimensional variability in traits, correct for sex-related measurement bias, and consider sex differences in the general population.

Finding the Phenotype

To date, the defining features of autism are based on clinical relevance and impairment. However, an autism phenotype simply describes a set of behaviors expressed due to underlying genetic difference; it does not need to relate to impairment or clinical relevance. Many behavioral traits appear to aggregate in families with autistic probands, which has been termed the broader autism phenotype (82). Characteristics include aloof and anxious personality traits, pragmatic language difficulties, motor coordination challenges, and cognitive deficits (language delay, difficulties with executive function, reading comprehension challenges) (82–85). Future studies should examine the distribution of traits thought to be related to underlying ASD genetics, including attentional processes, cognitive flexibility, sensorimotor control, temperament, and emotion regulation for consideration as part of the phenotypic definition for studies looking for underlying pathogenetic factors (82,85–87). Currently, one study has examined genetic variation and ASD outcomes in HL samples and found both copy number variants and polygenic risk scores underlying ASD diagnoses (88). Additional large-scale investigations that examine other traits as outcomes could also inform the autism phenotype.

Considering Developmental Course

Most studies of sex differences in ASD have used cross-sectional data. Studies that have examined change in autism symptoms over time have consistently found more reductions in ASD symptom severity in females compared with males, namely in social communication (89–91). In Burrows et al. (3), sex differences emerged early in development and were not specific to ASD. This is contrary to many studies that did not find sex differences in early development in ASD (8,92). It may be that a longitudinal study allows for identification of milder differences that persist over time. In a cross-sectional study, small differences may not be perceptible.

The timing of behavioral manifestation may also reflect differences in underlying biology. Autistic females are more likely to show profiles of reduced autism symptoms over time (89,90) and are also more likely to be diagnosed later in development (93,94). Autism in individuals who show later onset of symptoms may be etiologically and biologically different from an early-onset presentation, despite equifinality in symptom presentation. Cross-sectional studies are ill-suited to answering questions that pertain to change in presentation over time, and longitudinal studies of children who fall along the fuzzy diagnostic boundary are needed to better characterize variability in symptom manifestation over time.

Clinical Implications

This shift in strategies to elucidate sex-specific autism phenotypes calls into question the nature of ASD and the use of the same criteria for males and females. While the focus of this article is on the behavioral autism phenotype, the research reviewed here also has implications for clinical diagnosis of ASD. Female-specific risks have been recognized in other health conditions [e.g., hazardous lifestyle choices leading to cancer (95)], markers [e.g., female-specific signs of heart attacks (96)], and differential symptom thresholds for males versus females [as in depression (97–99)]. Similarly, the field of ASD should consider adjusting thresholds and symptom profiles for diagnosing ASD as well as sex-specific symptom manifestation. For many traits, including social communication and RRBs, sex differences exist in typically developing samples (11,15,91). Thus, it may be worth considering sex-specific deviations from average and nuanced expressions of autistic traits. Furthermore, the same level of autism traits may confer different impairment for males versus females given sex differences in social expectations and interaction styles. Measurement models that correct for sex-related biases could provide a framework to consider relative deviations from average.

Identifying sex-specific indicators of ASD liability could identify female-specific markers that could be incorporated into screening instruments and diagnostic tools. One such approach might look like applying a measurement framework to a set of behavioral indicators of an ASD phenotype and testing for loading-level differential item functioning by sex. If an item shows higher loading for females, it may be a stronger indicator of ASD risk and could thus be weighted more heavily. Here again, HL samples provide a novel opportunity to study very early behavioral indicators of ASD phenotypes for females. In Burrows et al. (3), we found that stereotyped and repetitive behavior was a stronger indicator of RRBs for females. Thus, when assessing for ASD, it may be worth weighting repetitive behaviors more heavily for females. It will be important to test whether similar sex bias exists beyond HL siblings to ensure generalizability, particularly given evidence of differences in HL samples (58).

There are likely to be traits that fall outside of DSM-defined autism symptoms that are relevant to underlying ASD genetics, particularly for females. It will be critical to identify traits that autistic girls (or HL siblings) show specific areas of difficulty in or differences compared with typically developing girls as candidate markers of the autism phenotype. Candidate traits that have been identified through family liability studies include attention regulation, cognitive flexibility, and emotion regulation. Biological models such as the female protective effect (100,101) and differing threshold for ASD in liability threshold (102,103) may help explain sex-related differences in expression of the ASD phenotype.

Studying individuals who fall at the boundary of diagnostic thresholds is inherently challenging, particularly when the current medical model relies so heavily on diagnosis for access to services. In Burrows et al. (3), we identified a higher proportion of females who fell in our high-concern group but who did not receive an ASD diagnosis at 24 months. It is worth highlighting the differences in the grouping methods: The high-concern group was identified based on changes in behavior from 6 to 60 months (rather than at a single time point), scores were corrected for measurement bias, and only one behavioral domain was considered (as opposed to a comprehensive clinical assessment). While not meeting full criteria for ASD, clinicians observed subclinical or related emotional dysregulation leading to a lack of clinical confidence in some cases. These children’s trajectories of social communication differed from typically developing children, and they may benefit from intervention. Future work should examine both early predictors of children who go on to have high social communication concerns, as well as later outcomes for these subgroups.

Significant variation exists in the space between ASD cases and controls. These children will not fit clearly within our diagnostic criteria, especially if we are only looking for a traditional definition of ASD. However, it may be helpful to consider ASD phenotypes and sex differences in their expression in terms of understanding the presentation of children who fall on this fuzzy boundary (104) (see Box 1). There is also likely to be significant diagnostic instability over time (90,105,106), particularly in boundary cases, necessitating repeated monitoring. Future research would benefit from examining heterogeneity in presentation over time and sex differences therein (Box 2). Health systems will continue to use the medical model to impose diagnostic dichotomies, but this approach is biologically implausible (2). Our science cannot continue to follow the same methods of requiring a diagnosis to warrant studying someone. Research methods that recruit children at varying liability for ASD and have variation in pleiotropic genes will help more to uncover the true nature of ASD trait manifestation and will likely result in sex ratios that are much lower than the 4:1 ratio that has long been established in ASD. The ASD phenotype may also extend beyond core ASD symptoms, including traits such as executive dysfunction, cognitive inflexibility, and emotion dysregulation, among others. If we can approach this work with epistemological curiosity and methodological rigor, we may identify ways to shift the medical system to better recognize the challenges associated with the autism phenotype, particularly for females. This shift in approach may help uncover subtypes of behavior that map more closely onto biology, open doors for improved identification of risk profiles, and tailor treatments that deliver the right dose of intervention to the right person at the right time.

Box 1. Clinical Vignette.

Noelle is a 12-year-old girl who has an older brother on the autism spectrum, and she participated in a study of children at high familial likelihood for ASD. Her early developmental profile was generally within the average range, with the exception of early motor delays based on parent report. At 24 months, her parents were not concerned with her development, although they endorsed minimal social chat, little interest in peers, and limited pretend play. No RRBs were endorsed per parent report. However, during the ADOS, she had meltdowns during transitions, particularly when she had to give up a preferred toy. Her eye contact was notably limited, and she had flat facial expressions, but she had some moments of nice, directed attention with the examiner. At several points during the visit, her mother attempted to engage her in play with some of the toys presented, but she frequently said “mine!” and turned away from her mother, which was clearly upsetting to her mother. She met criteria for pervasive developmental disorder but not full autistic disorder according to DSM-IV criteria. The clinician’s comments from this visit stated, “very confusing child; lots of autism symptoms, but no repetitive behavior so doesn’t meet criteria.” When she returned for her school age visit at 12, her parents described her as “creative, smart and inquisitive,” and their main concerns were that she was “moody.” She was an avid reader and described that she often liked to spend time alone, spacing out about her favorite books. Her cognitive scores were average, while her reading scores were more than 2 standard deviations above the mean. Her vocabulary was expansive, and she frequently provided lengthy responses to questions and elaborated on small details. On the ADOS, she had notably poor eye contact and poor insight into social relationships. She described having several friends, who she initially met because they were both “loners.” She only rarely saw them outside of school. Again, no RRBs were endorsed per parent report. She did not meet criteria for ASD in DSM-5, although clinicians reviewing the case felt that she had some clear social deficits and quirks. This represents a child with some clear genetic liability for ASD who has fallen along the fuzzy diagnostic boundary for much of her life.

ADOS, Autism Diagnostic Observation Schedule; ASD, autism spectrum disorder; RRB, restricted and repetitive behavior.

Box 2. Questions for Future Research.

• Why are we missing ASD in females at early ages? Do females show symptoms on the same developmental time course as males and our current ascertainment process is not picking up their concerns, or do females show a different trajectory of autism symptom onset?

• Are there sex-specific genes that contribute to the risk for and progression of ASD?

• Which biological models help explain sex differences in the autism phenotype?

• Are there sexually dimorphic brain regions and brain-behavior associations?

• Do we need different sets of criteria and cutoffs for females compared with males? How would we establish that need?

• How can we shift the current ascertainment, diagnosis, and intervention system to more comprehensively identify and address the range of concerns that present in females?

• What would intervention look like for females who do not fit the traditional diagnostic criteria if we are able to identify them early?

• What interventions would be most helpful for late-diagnosed females and males?

ASD, autism spectrum disorder.