Psychometric Evaluation of Two Adult Autism Screening Tools in Brazil
ABSTRACT
Despite increasing rates of autism spectrum disorder (ASD) in many countries, substantial evidence suggests persistent underdiagnosis of ASD in many low and middle-income countries, such as Brazil. Underdiagnosis or misdiagnosis may be particularly prevalent among women who often present subtler social and behavioral characteristics, engage in camouflaging strategies, and exhibit different symptom expressions compared to men. This study evaluates two new instruments to improve screening for ASD among adults in Brazil: the Screening for Autism in Adults (SfA-A) and the Screening for Autism in Females (SfA-F). A sample of 3302 Brazilian adults (mean age = 37.55 ± 11.34 years) completed the SfA-A, while 7738 Brazilian adult women (mean age = 38.77 ± 10.28 years) completed the SfA-F. Exploratory Structural Equation Modeling was conducted. Reliability was assessed using Cronbach's alpha, McDonald's omega, and test–retest. Criterion validity was determined by the AQ-10 and two autism-related questions. Norms were established based on percentiles. Exploratory and confirmatory factor analyses indicated that the SfA-A and SfA-F exhibited strong model fit, high internal consistency (α > 0.8), and initial evidence of criterion-related validity. The SfA-A and SfA-F were developed to address critical gaps in ASD screening among adults in Brazil. These tools hold promise for identifying ASD symptoms and can be used to initiate formal ASD evaluation.
Summary
- The study developed two new tools to identify autism in adults.
- One of the tools is specifically designed for women, addressing the unique characteristics of autism such as camouflage symptoms.
- Both tools demonstrated high accuracy and reliability.
1 Introduction
Autism spectrum disorder (ASD) is a complex neurodevelopmental condition, the main features of which involve persistent deficits in social communication and social interaction, repetitive patterns of behavior, and restricted interests or activities (American Psychiatric Association 2023; Lord et al. 2018). These symptoms often lead to significant impairments in daily functioning, affecting educational attainment, employment, social relationships, and quality of life. Individuals with ASD may also experience co-occurring conditions such as anxiety, depression, and sensory processing issues, which further exacerbate challenges and require comprehensive, multidisciplinary intervention strategies (Lai et al. 2014; Lord et al. 2018).
Historically, males have been overrepresented in this category by a ratio of 4:1 (Cook et al. 2021; Lai et al. 2011), although other findings show an increase in the prevalence of ASD among adult women (Rutherford et al. 2016; Wilson et al. 2016). Within Brazil, current estimates suggest that approximately 1.5 million individuals have received an ASD diagnosis, but recent estimates using base rates from other countries suggest that there may be as many as 4.5 million undiagnosed cases of ASD in Brazil (Paiva Jr 2023). Several recent studies suggest that this underdiagnosis is due to several factors, including the stigma associated with identification (Araujo et al. 2024), a lack of access to services within Brazil (Paula et al. 2020), a lack of awareness and training among professionals pertaining to the recognition of ASD symptoms (Bordini et al. 2014), and the recency of legal protections for individuals with ASD (Rios and Costa Andrada 2015). Each of these factors is thought to contribute to underdiagnosis as well as long wait times between initial suspicion of ASD and the ability to actually receive diagnosis within Brazil (Ndindeng 2024; Ribeiro et al. 2017). Thus, the ability to identify initial symptoms, along with some estimate regarding the severity of symptoms, is critical for supporting individuals, families, educators, and health care providers in identifying initial concerns.
Adults diagnosed later in life often experience a mix of relief and frustration—relief from finally understanding their lifelong struggles but also regret related to missed opportunities for early support that could have potentially improved their quality of life. These individuals frequently report challenges in forming and maintaining social relationships, securing stable employment, and developing a cohesive sense of identity, leading to feelings of isolation (Bargiela et al. 2016). Thus, the need for such tools in low- and middle-income countries is particularly important, where recent estimates suggest that less than 25% of the individuals who may actually qualify for an ASD diagnosis actually receive one (Huang et al. 2020; Ndindeng 2024; Paiva Jr 2023; Sukiennik et al. 2022).
The limited availability of standardized and culturally adapted screening tools for use with adult populations in Brazil has been well documented (Sukiennik et al. 2022). Current efforts rely on adapted versions of internationally recognized tools, such as the Autism Diagnostic Observation Schedule (ADOS; Pacífico et al. 2019) the Autism Diagnostic Interview-Revised (ADI-R; Becker et al. 2012), and the Childhood Autism Rating Scale (CARS, Pereira et al. 2008). However, there is a persistent need to develop and validate culturally grounded ASD screening tools for use with adult populations in Brazil, in part because cross-cultural differences can influence the expression and perception of ASD traits, but also to address the widespread underdiagnosis of adults within Brazil (Backes et al. 2014; Grinker 2007; Huang et al. 2020; Mandell and Novak 2005).
There are several reasons why it may be particularly important to develop such measures specifically for use with women. Research indicates that autistic traits in girls and women are more likely to present in less typical ways, which can result in underdiagnosis or delayed diagnoses (Cook et al. 2024). Girls and women who potentially have ASD also engage in higher levels of social camouflaging leading to fewer observable symptoms during standard clinical assessments (Hull et al. 2020; Ratto et al. 2018). Camouflaging behaviors, such as mimicking social interactions, suppressing repetitive behaviors, and adapting communication styles often lead to misdiagnosis or delayed diagnosis in women (Rynkiewicz et al. 2016; Lai and Baron-Cohen 2015). Additional evidence suggests that diagnostic criteria for ASD are based on male-dominated samples and thus may overlook manifestations more common in females, such as subtle social difficulties or internalized distress (Lai et al. 2014). Thus, the ability to assess more subtle manifestations of ASD can help to close a longstanding diagnostic gap.
In the current study, we developed two separate self-report screening instruments. The choice to rely on self-report was based on several considerations. First, traditional diagnostic approaches often rely on developmental history and third-party observations, which may not be readily available for older individuals (Baron-Cohen et al. 2001; Pagán et al. 2023). Second, adults generally develop greater self-awareness and insight into their own experiences, making self-report measures valuable for identifying lifelong autistic traits that may not be captured through clinical observation alone (Arrow et al. 2023; Keith et al. 2019). Third, self-report tools offer a more efficient and accessible approach to screening, thus reducing the need for more resource-intensive approaches such as clinician-administered evaluations (Brentani et al. 2021).
Despite the advantages of self-report screening measures, there is currently a dearth of validated ASD screening tools for non-English-speaking populations (Sukiennik et al. 2022). In Brazil, the lack of culturally adapted instruments further limits access to early identification and intervention, particularly among women (Paula et al. 2020). This is unfortunate because delayed access to appropriate interventions can increase emotional exhaustion and higher rates of mental health issues among undiagnosed adults (Hull, Petrides, et al. 2017; Lai and Baron-Cohen 2015). Thus, continuing to develop and evaluate culturally and gender-sensitive self-report screening tools is an important step for ensuring greater access to formal diagnosis and treatment (Wigham et al. 2019).
2 Current Study
The current study had two primary aims. The first was to evaluate the psychometric properties of a newly developed screening measure to assess initial symptoms of ASD among adults, regardless of birth sex. The need for self-report screening tools for use with adults is underscored by challenges associated with late diagnosis, the reliance on subjective experiences, and the lack of measures specifically validated for diverse adult populations (Wigham et al. 2019). Such tools have the potential to provide initial information regarding ASD traits that may have been overlooked, particularly in individuals with subtle or atypical symptom presentations. Based on the unique characteristics and potentially detrimental effects of delayed diagnosis among women (Hull, Petrides, et al. 2017; Lai and Baron-Cohen 2015), our second aim was to evaluate the psychometric properties of a measure designed specifically to screen for ASD among women. We anticipated that both measures would demonstrate adequate psychometric properties, including strong internal consistency, model fit, and criterion-related validity.
3 Methods
3.1 Sample
In Table 1 we present the demographic characteristics of both samples. The first sample consisted of 3302 participants who responded to a general questionnaire and the second included 7738 female participants who completed a questionnaire tailored specifically for women. The mean age of the general sample was 37.55 years (SD = 11.34), with ages ranging from 18 to 78 years old; 77.4% identified as female, 19.1% as male, and 3.5% as “other”; and 44.9% had completed an undergraduate degree, 38.2% a graduate degree, 11.5% reported high school, elementary school (1.9%), or other forms of education (3.5%). Participants were also asked about clinical diagnoses. In the general sample, 22.7% reported having a diagnosis of ASD, 18.9% ADHD, and 10.3% had OCD.
| Characteristic | General (N = 3302) | Females (N = 7738) |
|---|---|---|
| Age | ||
| Mean (SD) | 37.55 (11.34) | 38.77 (10.28) |
| Range | 18.00–78.00 | 18.00–89.00 |
| Civil status | ||
| Single | 1381 (41.8%) | 2855 (36.9%) |
| Married | 1089 (33.0%) | 2687 (34.7%) |
| Stable union | 463 (14.0%) | 1288 (16.6%) |
| Divorced | 329 (10.0%) | 810 (10.5%) |
| Widower | 40 (1.2%) | 98 (1.3%) |
| Gender | ||
| Female | 2555 (77.4%) | 7738 (100.0%) |
| Male | 631 (19.1%) | — |
| Other | 116 (3.5%) | — |
| Level of education | ||
| Elementary school | 63 (1.9%) | 89 (1.2%) |
| High school | 379 (11.5%) | 688 (8.9%) |
| Undergraduate | 1481 (44.9%) | 3092 (40.0%) |
| Graduate | 1263 (38.2%) | 3599 (46.5%) |
| Other | 116 (3.5%) | 270 (3.5%) |
| Do you have a diagnosis of autism spectrum disorder (ASD)? | ||
| Yes | 750 (22.7%) | 1091 (13.7%) |
| No/other response | 2114 (77.3%) | 6867 (86.3%) |
| Do you have a diagnosis of Attention-deficit (ADHD)? | ||
| Yes | 636 (18.9%) | 1586 (19.9%) |
| No/other response | 2721 (81.1%) | 6372 (80.1%) |
| Do you have a diagnosis of Obsessive-Compulsive Disorder? | ||
| Yes | 349 (10.3%) | 828 (10.4%) |
| No/other response | 3008 (89.7%) | 7130 (89.6%) |
The mean age of the female-only sample was slightly higher (M = 38.77, SD = 10.28), with a greater range (i.e., 18–89 years old). Participants in the general sample reported being single (41.8%), married (33.0%), in a stable union (14.0%), divorced (10.0%), or widowed (1.2%). Among the female-only sample, 36.9% reported being single, 34.7% married, 16.6% in a stable union, 10.5% divorced, and 1.3% widowed; 46.5% reported graduate-level education, which was higher than the general sample, with distributions in the other categories similar to those reported in sample 1. Among the female sample, the prevalence of ASD was 13.7%, while self-reported rates of ADHD (19.9%) and OCD (10.4%) were similar to those reported in the general sample.
3.2 Instruments
3.2.1 Sociodemographic Measure
Before completing the study instruments, participants in both samples answered questions about their sociodemographic characteristics, including gender (male, female, or other; coded as 1 = female, 2 = male, and 3 = other), current age, marital status (coded as 1 = single, 2 = married, 3 = stable union, 4 = divorced, or 5 = widowed), and highest level of education achieved (coded as 1 = elementary, 2 = high school, 3 = undergraduate, 4 = graduate, and 5 = other). Participants were also asked whether they had been diagnosed with ASD, ADHD, or OCD and whether they considered themselves to have any mental condition or disorder (either self-identified or professionally diagnosed). For those answering “yes” to the latter question, an open-ended text box was provided for them to elaborate on their mental health status. Responses were categorized as ‘no diagnosis’ (coded = 0) and as ‘yes’ (coded = 1). Participants who left this field blank or provided unclear responses were coded “0”. Responses that included descriptive text but did not confirm a diagnosis were grouped under ‘other’.
3.2.2 Screening for Autism in Adults
The screening for autism in adults (SfA-A) is a 75-item self-report measure developed to assess autistic traits in adults. Item development was based on DSM-5-TR diagnostic criteria, an extensive literature review, and expert evaluations to ensure that the items are both clear and relevant for identifying ASD symptoms (Benevides et al. 2024). Items evaluate key domains of ASD including social communication challenges (e.g., “I laugh or smile at the wrong time”), difficulties in establishing and maintaining relationships (e.g., “I find it difficult to maintain relationships”), routinized behaviors (e.g., “I feel a great need for routine on a daily basis”), and sensory sensitivity [e.g., “I find that I am more sensitive to sensory stimuli (sounds, lights, and touches) than most people”]. Responses are provided on a 5-point Likert scale ranging from 1 = Strongly Disagree to 5 = Strongly Agree. In the initial development study (Benevides et al. 2024), content validity was assessed using the Content Validity Coefficient (CVC), with a cutoff of 0.7 to determine adequacy. The highest-rated items for clarity included “have the habit of rehearsing speeches or behaviors in front of the mirror to later use in social situations” which received a CVC clarity score of 1.0 and a CVC relevance score of 0.92. The average CVC for all items exceeded 0.7, though some items scoring slightly below this threshold were retained due to their alignment with DSM-5-TR criteria.
3.2.3 The Screening for Autism in Females
The screening for autism in females (SfA-F) is a 34-item self-report measure designed specifically to assess women (Benevides et al. 2024). The selection of items was based on three key sources: (a) adaptation of screening questionnaire items for the general adult population, such as the AQ-10 and Aspie-Quiz; (b) constructs from international scales targeting the female ASD traits (e.g., GQ-ASQ, Attwood et al. 2020; CAT-Q, Hull et al. 2019); and (c) newly created items that were designed specifically to assess camouflaging behaviors and gender-related interests.
Although the SfA-F is not directly based on DSM-5-TR diagnostic criteria, it incorporates one factor—sensory sensitivity—that aligns with the DSM-5-TR's domain of restricted and repetitive behaviors. This factor captures heightened sensory reactivity, including intense sensitivity to textures, sounds, and physical contact, which are frequently reported among autistic individuals (e.g., “Everyday situations, such as cutting nails/hair or being touched unexpectedly, cause me discomfort or distress,” “I feel distress when grooming myself [e.g., combing or cutting hair, nails] or being touched [e.g., someone touching my feet]”). The remaining items capture symptoms associated with camouflaging and gender-related experiences, which have been identified as key characteristics of the female autistic phenotype but are often underrepresented in traditional diagnostic frameworks. For instance, camouflaging behaviors involve consciously mimicking social behaviors, facial expressions, and speech patterns to fit in with peers (e.g., “I often copy or imitate the social behavior of other women” and “When I interact with someone, I strive to mimic their facial expressions and body language”). Other items focused on masking behaviors, such as suppressing autistic traits and adopting a “social mask” to appear more neurotypical (e.g., “I often feel like I need to hide my true self in order to be socially accepted” and “In social situations, I feel like I have to fake it to look ‘normal”). Responses are provided on a 5-point Likert scale, ranging from 1 = Strongly Disagree to 5 = Strongly Agree. Similar to the SfA-A, the SfA-F demonstrated adequate content validity, with an average content validity coefficient (CVC) above 0.7 in expert reviews (Benevides et al. 2024).
3.2.4 Autism Spectrum Quotient-10
Participants in both samples also completed the widely used autism spectrum quotient-10 (AQ-10). The AQ-10 was originally developed from the full 50-item Autism Spectrum Quotient (Baron-Cohen et al. 2001) and includes 10 items (e.g., “I often notice small sounds when others do not”) and “I like to collect information about categories of things (e.g., types of car, types of bird, types of train, types of plant etc.)”. Responses are provided on a 4-point scale ranging from “Definitely Agree” to “Definitely Disagree” and responses are scored as “1” for “Definitely agree” or “Slightly agree” responses and “0” for Disagree or Strongly Disagree. Scores at or above 6 suggest possible autism and the need for a full diagnostic assessment. The tool is recommended by the National Institute for Health and Care Excellence to help front line clinicians and social care professionals screen for ASD traits efficiently. Recent studies using item response theory demonstrate that the use of the AQ-10 sum score in its current form is unbiased by gender (Murray et al. 2014). However, concerns that the AQ-10 may not be psychometrically sound in non-clinical samples have been raised by researchers (c.f., Taylor et al. 2020). In the current study, the alpha coefficients for both samples were 0.74 and 0.72, respectively.
3.3 Procedures
This study was approved by the Brazilian Research Ethics Committee (Plataforma Brazil, Protocol number: 71169423.3.0000.5589). To participate, individuals had to be residents of Brazil and 18 years or older. Those under 18 or not residing in Brazil were excluded from participation. Participants were recruited in 2023 through social media posts on Instagram, using a profile dedicated to individuals with ASD. All posts contained a brief study description, consent forms, and a direct link to the questionnaires, all of which were hosted on Google Forms. Additionally, all participants were asked to share the link with their networks, including families and friends. Although this non-probability sampling approach can compromise generalizability, this “snowball sampling” technique is particularly valuable for recruiting individuals who have traits or characteristics that occur less frequently in the population (Lee and Spratling 2019; Soda et al. 2021). After agreeing to participate, participants then proceeded with the questionnaires, which included either the SfA-A or SfA-F, the AQ-10, and self-reports of sociodemographic data.
3.4 Analytic Plan
Prior to conducting any analyses, data were transferred to R (Version 4), Python (Version 3.11), and Mplus (Version 8) (Muthén and Muthén 2017). In an effort to provide initial validation of our results, we examined internal structure, relationships with other variables, and test consequences (American Educational Research Association et al. 2014). First, to explore the underlying factor structure of both tools, we conducted iterative exploratory structural equation modeling (ESEM) on a randomly selected half of each dataset. The ESEM utilized an oblique rotation with diagonally weighted least squares (WLSMV) estimation, a technique specifically designed for ordinal data. The number of factors was determined by a combination of statistical fit indices and theoretical considerations. Once the ESEM was completed, confirmatory factor analysis (CFA) was then conducted on the remaining half of each dataset.
Final models were selected based on statistical fit indices, theoretical interpretability, and factor loadings. Comparative fit index (CFI) and Tucker-Lewis index (TLI) values greater than 0.90 and root mean square error of approximation (RMSEA values less than 0.08 were considered indicative of good model fit; Xia and Yang 2018). A minimum factor loading of 0.32 was required for inclusion in the final model (Tabachnick and Fidell 2018). Factors were named based on item loadings and theoretical relevance to ASD symptoms. Once factors were established, internal consistency was evaluated using Cronbach's alpha and McDonald's omega coefficients.
The evidence based on relations to other variables, Criterion-related validity, was evaluated in three ways. First, we examined the correlation between scores on the AQ-10 and scores on the SfA-A and SfA-F. Second, independent t-tests comparing mean differences on the SfA-A and SfA-F for groups above and below the recommended AQ-10 cutoff scores were conducted. Third, we grouped participants based on two self-report items that asked “Have you ever been diagnosed with autism” and “Do you suspect you are on the spectrum, even without a formal diagnosis” and then compared these groups on both measures. Lastly, to enhance the initial practical utility of the tools, a preliminary normative table using percentile ranks and qualitative indicators was developed. All data, R and Python codes, and Mplus syntax are available at https://osf.io/gqmkd/.
4 Results
4.1 Internal Structure
Prior to performing ESEM factor analysis on a randomly selected half of each dataset (i.e., SfA-A and SfA-F) we first confirmed that items were not linearly dependent. Results of the ESEM on both samples indicated that the optimal model fit was achieved with a four-factor solution. In the analysis involving the SfA-A, the four-factor solution explained 58.3% of the variance and factor labeling included: “Social Communication” (22 items), “Social Interaction” (17 items), “Sensory Sensitivity” (24 items), and “Repetitive Patterns of Behavior, Interests, or Activities” (12 items). For the SfA-F, the solution accounted for 61.4% of variance and factors were labeled: “Camouflage Compensation” (10 items), “Gender-Related Interests” (4 items), “Camouflage Masking” (16 items), and “Sensory Sensitivity” (4 items).
4.2 Confirmatory Models
Following the results of the ESEM, two CFAs were conducted to verify the factor structures on the remaining half of each sample. An overview of these results is presented in Tables S1 and S2. For the SfA-A, the CFA demonstrated strong model fit, [χ2(2694) = 70745.449, p < 0.001, CFI = 0.975, TLI = 0.974, RMSEA = 0.098, SRMR = 0.046]. The internal consistency of these factors was high: Social Communication (α = 0.95, ω = 0.96), Social Interaction (α = 0.95, ω = 0.96), Sensory Sensitivity (α = 0.95, ω = 0.96), and Repetitive Patterns of Behavior, Interests, or Activities (α = 0.92, ω = 0.93). Notably, however, the RMSEA in this analysis was above recommended levels, which may have been due to the large sample size and the reliance on ordinal data, which can lead to high RMSEA values even when the model fits well (Kenny et al. 2015; Xia and Yang 2018). However, the specific limitations of unbiased RMSEA in such contexts are not fully detailed in the available literature, suggesting a need for further research on its application with ordinal data.
Intercorrelations between factors ranged from moderate to large. Social Communication was associated with Social Interaction (r = 0.823), Sensory Sensitivity (r = 0.840) and Repetitive Patterns of Behavior, Interests, or Activities (r = 0.753). Social Interaction was associated with Sensory Sensitivity (r = 0.792) and Repetitive Patterns of Behavior, Interests, or Activities (r = 0.770). The correlation between Sensory Sensitivity and Repetitive Patterns of Behavior, Interests, or Activities was (r = 0.808).
The CFA results for the SfA-F are provided in Table S2. Results indicated an adequate fit [χ2 (521) = 30483.428, CFI = 0.987, TLI = 0.986, RMSEA = 0.086, SRMR = 0.040] and the internal consistency reliabilities on each factor were strong: Camouflage-Compensation (α = 0.93, ω = 0.94), Gender-Related Interests (α = 0.79, ω = 0.88), Camouflage-Masking (α = 0.94, ω = 0.94), and Sensory Sensitivity (α = 0.81, ω = 0.84). Intercorrelations between factors ranged from moderate to large (i.e., Camouflage-Compensation and Gender-Related Interests, r = 0.525; Camouflaging Masking r = 0.889, and Sensory Sensitivity r = 0.753, respectively). Gender-Related Interests was associated with both Camouflaging Masking (r = 0.577) and Sensory Sensitivity (r = 0.595). The correlation between Camouflaging Masking and Sensory Sensitivity was r = 0.821.
4.3 Criterion-Related Validity
To evaluate initial criterion-related validity of each measure, we next examined associations between responses on the AQ-10 and responses on the SASfA-A. The correlation between the total scores on the SfA-A and AQ-10 (r = 0.73, p < 0.001) and between the SfA-F and AQ-10 (r = 0.68, p < 0.001) were both large. Next, using AQ-10 guidelines (Allison et al. 2012) we classified individuals as either meeting (6 or greater) or not meeting the AQ-10 criteria for suspected autism, and then compared these groups on SfA-A and SfA-F total scores. Participants meeting or exceeding the AQ-10 criterion had significantly greater scores on the SfA-A (M = 302.58, SD = 38.15) than did those below AQ-10 guidelines (M = 229.61, SD = 56), t(1889) = 39.176, p < 0.001. Similarly, in the SfA-F sample, the group above the AQ-10 criterion had significantly greater scores on the SfA-F (M = 136.12, SD = 18.89) than did those below the cutoff (M = 104.13, SD = 27.98), t(6686.5) = 56.136, p < 0.001. Figure 1 provides a visual depiction of the characteristics of each distribution.
Lastly, using self-report data, participants who answered “yes” to the item “Have you ever been diagnosed with autism?” had significantly greater scores on the SfA-A ((M = 314.44, SD = 32.59) than did individuals who responded “no,” (M = 261.25, SD = 59.72), t(2299.5) = 31.714, p < 0.001). Similarly, those who answered “yes” to the same item on the SfA-F had significantly greater scores (M = 145.18, SD = 13.19) than did those who indicated “no” on this item (M = 116.04, SD = 28.88), t(215341) = 48.942, p < 0.001. Participants who reported “suspecting they had ASD” had significantly greater scores on both the SfA-A (M = 286.36, SD = 40.19 vs. M = 267.85, SD = 64.67; t(2842.4) = 9.962, p < 0.001) and the SfA-F (M = 130.28, SD = 19.6 vs. M = 114.48, SD = 31.38, t(6245.8) = 25.116, p < 0.001) than did those who did not suspect ASD.
4.4 Proposed Norms
Because the distribution of both tools was non-symmetrical, we calculated percentile ranking scores. Percentiles are ideal for non-symmetrical distributions as they provide a more accurate view of data spread and central tendency than does the mean and standard deviation. Unlike the mean, percentiles are robust to outliers and are less affected by extreme values. Percentiles also do not rely on assumptions of normality, making them suitable for non-parametric analyses. Table 2 shows the distribution of scores for the SfA-A. The scores are presented across percentiles (1–99), with higher values reflecting greater symptom severity. Table 3 presents the percentile distribution of scores for the SfA-F.
| Percentile | Overall | Domains | |||
|---|---|---|---|---|---|
| Social communication | Social interaction | Sensory sensitivity | Repetitive | ||
| 1 | 76 | 22 | 17 | 24 | 12 |
| 5 | 157 | 38 | 34 | 45 | 27 |
| 10 | 191 | 47 | 43 | 55 | 32 |
| 15 | 210 | 54 | 49 | 62 | 36 |
| 20 | 226 | 58 | 54 | 68 | 39 |
| 25 | 240 | 62 | 58 | 73 | 42 |
| 30 | 251 | 66 | 62 | 76 | 43 |
| 35 | 259 | 69 | 64 | 80 | 45 |
| 40 | 268 | 72 | 66 | 84 | 47 |
| 45 | 275 | 75 | 68 | 87 | 48 |
| 50 | 284 | 78 | 70 | 90 | 49 |
| 55 | 291 | 81 | 71 | 93 | 50 |
| 60 | 298 | 83 | 73 | 96 | 51 |
| 65 | 305 | 85 | 74 | 98 | 53 |
| 70 | 311 | 88 | 76 | 101 | 54 |
| 75 | 318 | 90 | 77 | 104 | 55 |
| 80 | 324 | 93 | 79 | 107 | 56 |
| 85 | 332 | 96 | 80 | 109 | 57 |
| 90 | 342 | 99 | 82 | 112 | 59 |
| 95 | 353 | 103 | 83 | 115 | 60 |
| 99 | 375 | 110 | 85 | 120 | 60 |
| M | 273.3 | 75.2 | 65.8 | 86.5 | 47.1 |
| SD | 59.1 | 19.6 | 15.2 | 21.7 | 10.1 |
| Percentile | Overall | Domains | |||
|---|---|---|---|---|---|
| Camouflage: compensation and assimilation | Gender-related interests | Camouflaging: masking | Sensory sensitivity | ||
| 1 | 34 | 10 | 4 | 16 | 4 |
| 5 | 62 | 14 | 4 | 33 | 6 |
| 10 | 77 | 18 | 6 | 41 | 8 |
| 15 | 87 | 20 | 7 | 47 | 10 |
| 20 | 95 | 23 | 7 | 51 | 11 |
| 25 | 102 | 25 | 8 | 55 | 11 |
| 30 | 107 | 27 | 9 | 57 | 12 |
| 35 | 112 | 29 | 9 | 60 | 13 |
| 40 | 117 | 31 | 10 | 61 | 14 |
| 45 | 121 | 32 | 11 | 63 | 15 |
| 50 | 125 | 34 | 11 | 65 | 15 |
| 55 | 128 | 35 | 12 | 66 | 16 |
| 60 | 132 | 37 | 13 | 68 | 16 |
| 65 | 135 | 38 | 13 | 69 | 17 |
| 70 | 139 | 39 | 14 | 71 | 17 |
| 75 | 142 | 41 | 15 | 72 | 18 |
| 80 | 146 | 42 | 16 | 74 | 18 |
| 85 | 149 | 44 | 16 | 75 | 19 |
| 90 | 153 | 46 | 17 | 77 | 19 |
| 95 | 159 | 48 | 19 | 78 | 20 |
| 99 | 170 | 50 | 20 | 80 | 20 |
| M | 119.6 | 32.7 | 11.4 | 61.7 | 14.4 |
| SD | 29.1 | 10.3 | 4.3 | 13.7 | 4.2 |
Although percentiles provide a flexible and robust method for interpreting scores, we acknowledge the need for provisional cutoff guidance to assist clinicians in making professional judgments. That said, we used the mean score of participants positively screened as potentially on the spectrum by the AQ-10 and compared their results on the SfA-A and SfA-F. Using this method, we established a provisional cutoff that reflects the average score of positively screened individuals, providing clinicians with a practical yet flexible guideline while acknowledging the need for further empirical validation, including further sensitivity and specificity studies. Based on this initial analysis, however, we propose that scores below the 60th percentile suggest a lower likelihood of ASD, whereas scores at or above the 60th percentile are suggestive of autistic traits, with risk increasing progressively to the 99th percentile.
5 Discussion
This study investigated the initial psychometric properties of two instruments designed for ASD screening with adults in Brazil. Both instruments demonstrated acceptable levels of reliability and validity. Our analyses revealed that the measures effectively captured the core characteristics of ASD, as reflected in significant factor loadings consistent with theoretical expectations, but the RMSEA values for the SfA-A were inflated which may have been due to the large sample size and ordinal data (Shi et al. 2020). Despite this issue, the extracted factors from the SfA-A and SfA-F were well-aligned with the core diagnostic domains of ASD as described in DSM-5-TR (American Psychiatric Association 2023) and ICD-11 (WHO, 2023). Items on the “Social Communication” and “Social Interaction” factors of the SfA-A correspond to persistent deficits in the social communication and interaction domains whereas items in the “Sensory Sensitivity” and “Repetitive Patterns of Behavior, Interests, or Activities” factors align with the restricted and repetitive behaviors.
For the SfA-F, items on the “Camouflaging-Compensation” and “Camouflaging-Masking” factors reflect mechanisms that contribute to atypical social communication, particularly among women who may have utilized strategies to hide or mask autistic behaviors (American Psychiatric Association 2023). The “Gender-Related Interests” factor provides additional insight into gender-related behavioral expressions of ASD, while the “Sensory Sensitivity” factor corresponds to sensory processing difficulties described within the restricted and repetitive behaviors domain of the DSM-5-TR.
While previous screening tools, such as the AQ-10 (Allison et al. 2012) and RAADS-R (Ritvo et al. 2011), have been widely used for ASD detection, these instruments were developed primarily in Western contexts and have not been adapted for Brazilian populations. As noted in one review, the most common approach used for cultural adaptation of screen and diagnostic tools is “surface modifications” such as simple translation of items into a new language (Al Maskari et al. 2018). By contrast, the SfA-A and SfA-F were developed specifically for the Brazilian population, incorporating local language, norms, and behaviors that are more relevant to the experiences of Brazilian adults with autism.
In addition to issues pertaining to cultural adaptation, the AQ-10 consists of only 10 items and may not fully capture the multidimensional nature of ASD, whereas the SfA-A was designed to provide a more comprehensive assessment of multiple domains to ensure that core and associated ASD traits are considered. The ability of the SfA-A to differentiate individuals with self-reported diagnoses or suspected ASD from those without such suspicions suggests that this is a promising tool for initial screening among adults in clinical and community settings within Brazil.
Whereas the SfA-A offers promise for screening ASD symptoms among all adults, the SfA-F was designed specifically to address camouflaging-related behaviors in recognition of gender differences in ASD expression. The derived factors on the SfA-F reflect adaptive strategies that are often overlooked by traditional screening tools, which can obscure ASD traits in females (Cook et al. 2021; Onaolapo and Onaolapo 2017). Given that the DSM-5-TR and ICD-11 emphasize gender differences in ASD presentation, the structure of the SfA-F provides a promising approach for addressing screening gaps among women by incorporating behaviors associated with camouflaging and social compensation (Dean et al. 2017; Hull et al. 2019). Integrating camouflaging behaviors and GenderRelated Interests into the SfA-F screening may have relevance for non-Western contexts where gender differences in diagnostic processes remain understudied (Sukiennik et al. 2022). Thus, although the ability to detect these subtle yet clinically relevant behaviors can help to ensure greater awareness regarding the specific needs and presentation of ASD symptoms among women, continuing to investigate the manifestation of these behaviors and their relation to accurate diagnosis will continue to be important in countries like Brazil where there are high rates of underdiagnosis.
5.1 Study Limitations
This study has several limitations that should be considered with the findings. First, data were gathered online using a social media-based recruitment strategy. While this approach allowed for a large and diverse sample, it raised the potential of selection bias as the study was shared in spaces frequented by individuals with an interest in neurodevelopmental conditions. Although online recruitment may have increased accessibility, it may have disproportionately reached urban, highly educated, and digitally connected individuals, limiting generalizability to populations with lower digital literacy or limited internet access (Paula et al. 2020). Future research should include more geographically and socioeconomically diverse participants, such as individuals from less connected regions, rural areas, or community-based clinical settings, to improve generalizability.
A second limitation is that we relied on subjective perceptions of behaviors and traits, which are susceptible to cognitive biases, symptom masking, and co-occurring conditions (Brugha et al. 2011; Cassidy et al. 2018). For example, individuals with high cognitive abilities may develop compensatory strategies that lead to underreporting of certain ASD traits (Hull, Petrides, et al. 2017) and conditions such as anxiety, depression, and ADHD frequently co-occur with ASD, potentially affecting responses in a self-report context (Lever and Geurts 2016; Sizoo et al. 2015). Although the SfA-A and SfA-F were designed specifically to assess ASD traits, the high rates of comorbid conditions (e.g., ADHD, OCD, and schizoid personality traits) highlight the need for further examination of discriminant validity. Future studies should employ differential item functioning (DIF) analysis to assess whether specific items disproportionately capture traits of other conditions rather than ASD itself. Additionally, item-level comparisons between individuals with ASD, ADHD, and OCD would help refine the specificity of these measures, ensuring that the SfA-A and SfA-F primarily assess autistic traits rather than broader psychopathological features. Such research would be enhanced by exploring alternative screening approaches such as behavioral assessments or machine-learning-based screening methods (Ruzich et al. 2015; Keith et al. 2019).
A third and related limitation is that we used self-report data regarding prior ASD diagnoses and suspected diagnosis as part of our analyses of the criterion-related validity of both measures. Although these analyses verified differences between groups in the anticipated direction, incorporating clinical confirmation of ASD diagnoses in future analyses with the SfA-A and SfA-F would help to strengthen the validity of these measures.
A fourth potential limitation is that the high number of items per domain has the potential to contribute to respondent fatigue and low completion rates (Adams et al. 2020; Lorenz and Heinitz 2014). Prior research suggests that longer screening tools may lead to reduced response quality, particularly among populations with attention and processing difficulties (Brugha et al. 2011; Cassidy et al. 2018). Given these challenges, future research should explore modifications of the SfA-A and SfA-F that could potentially improve usability and feasibility such as shortening the tool, developing an adaptive screening format, or validating briefer versions (Ruzich et al. 2015; Sizoo et al. 2015). By incorporating adaptive screening methods or shorter, high-sensitivity versions, future iterations of these tools could improve accessibility and effectiveness, ensuring they remain practical for clinical, research, and self-assessment purposes (Brentani et al. 2021).
5.2 Conclusion and Future Directions
This study introduced and validated two autism screening tools for adults in Brazil to address the persistent underdiagnosis of ASD in Brazil and critical gaps in ASD assessment. The SfA-A provides a comprehensive screening instrument for adults, while the SfA-F offers a gender-sensitive tool that accounts for social camouflaging and other female-specific autistic traits. While these newly developed measures provide an important contribution to adult ASD screening in Brazil, additional research on these and other similar tools is needed to further validate the findings presented here, to further refine cutoff scores, and to explore the feasibility of shorter versions or adaptive screening methods. Exploring digital adaptations and remote screening applications could also enhance accessibility, especially for underdiagnosed and underserved populations. Developing and testing online or mobile-friendly versions of these tools could facilitate large-scale screening efforts and improve access to early identification and intervention. Cross-cultural validation in other Portuguese-speaking countries would also be beneficial, expanding the applicability of these measures beyond Brazil (Paula et al. 2020). By addressing these areas, future research can further establish the robustness and practical applicability of the SfA-A and SfA-F to ensure they facilitate greater access to evaluation and treatment services among adults in clinical and community settings.
Ethics Statement
Dr. Luis Anunciação served as the principal investigator of this study, which received approval from the Pontifical Catholic University of Rio de Janeiro as well as from the Brazilian Research Ethics Committee (Plataforma Brasil, Protocol number: 71169423.3.0000.5589). Recruitment took place in 2023 through social media, where participants accessed either the SfA-A or SfA-F questionnaires via links to Google Forms. To qualify for participation, individuals had to meet the criteria: (1) be residents of Brazil, (2) be 18 years old or older, and (3) agree to the research consent form. Participants who were under 18 or residing outside Brazil were not incuded in analyses. No financial compensation was offered for participation.
Conflicts of Interest
The authors declare no conflicts of interest.
Open Research
Data Availability Statement
The data that support the findings of this study are openly available in open science framework at https://osf.io/gqmkd/.
