Sex as a Moderator Between Parent Ratings of Executive Dysfunction and Social Difficulties in Children and Adolescents with Autism Spectrum Disorder

Participants were recruited from Norwegian health services specializing in the assessment of ASD and other neurodevelopmental disorders. The study was part of the national BUPGEN network (Grove et al., 2019) and the methods are described previously in Torske et al. (2017). The current sample consisted of 25 girls and 91 boys with ASD who were recruited between 2013 and May 2018 and assessed at age 5–19 years. Fifteen of the children (2 girls, 13 boys) were diagnosed with childhood autism, 9 (2 girls, 7 boys) with atypical autism, 57 (14 girls, 43 boys) with Asperger syndrome and 35 (7 girls, 28 boys) with unspecified pervasive developmental disorder (PDD-NOS). Because ASD and ADHD often co-occur (Lord et al., 2018), the current study also included children with ASD and comorbid ADHD.

The male:female ratio was 3.6:1. In total, 40 children (34.5%) had a comorbid disorder of ADHD, and the male:female ratio was 7:1 in the ADHD group. All participants had a full-scale intelligence quotient (IQ) ≥ 70 based on a standardized Wechsler’s test (Full-scale IQ). We did not use any standardized assessment to determine if the participants spoke Norwegian fluently. However, the psychologist who administered the intelligence test carefully evaluated the language fluency of the children in relation to the test session and made sure that they understood the instructions for the test.

Participants were assessed by a team of experienced clinicians (clinical psychologists and/or child psychiatrists and educational therapists). Diagnostic conclusions were best-estimate clinical diagnoses derived from tests, interview results, and observations. All diagnoses were based on the International statistical classification of diseases and related health problems 10th revision (ICD-10) (World Health Organization, 1992) criteria, and the autistic symptoms were evaluated using the autism diagnostic observation schedule (ADOS-G and ADOS-2) modules 2, 3 or 4 (Lord et al., 2000, 2012) and/or autism diagnostic interview-revised (ADI-R) (Rutter et al., 2003). In addition, the assessment included a full medical and developmental history, physical examination, IQ assessment and parent ratings of EF. The ADHD diagnosis were based on a thorough diagnostic process including detailed clinical examination, cognitive tests, and questionnaires. However, unfortunately, we did not have access to severity score of ADHD symptomatology for this sample.

Analyses were conducted using the R statistical environment (version 4.2.0) using the “jmv” (Version 2.3.4; (Selker et al., 2017)). The FUZZY extension command in SPSS was used to generate a closely matched sample of boys and girls on age and IQ. Checks for multicollinearity, homoscedasticity, influential outliers, homogeneity of variance, and normality of residuals suggested that assumptions for analyses were met. Statistical significance was set at p < 0.05 and adjusted according to number of comparisons. We provide justifications below for how we adjusted tests for multiple comparisons to control the Type-I error rate. Conventional values were used for interpreting effect sizes (Effect size values of 0.2, 0.5, and 0.8, were considered small, medium, and large effects, respectively (Cohen, 1988)).

We have used T-scores from the BRIEF in all the analyses. T-scores can provide clinical meaning to scores, since age and sex can affect what is considered “typical” behavior. Furthermore, we find it important to use t-scores because deviations from TD is important, particularly as girls and boys develop in different ways and parents might have different expectations between girls and boys. In general, age influences t-scores more than sex differences. For example, a t-score of 69 on GEC corresponds to a raw score of 163–164 for girls aged 8–10 years and a raw score of 165–166 for boys aged 8–10 years.

Welch’s t-tests were conducted to investigate sex differences in ADI-R and BRIEF scores. We used the Welch’s t-test because it is recommended as a default instead of the Student´s t-test, even if variances are equal (Delacre et al., 2017). As we were examining a series of tests and hypothesizing that these groups were not significantly different, we adjusted for 6 tests, ADI-R A, ADI-R B, ADI-R C, BRIEF GEC, MI and BRI (critical p-value = 0.008); (Armstrong, 2014; Perneger, 1998), with values less than 0.05 considered on the border of statistical significance (i.e., Bonferroni correction). A chi-squared statistic was calculated to assess the frequency distribution of comorbid ADHD between sexes. For the t-tests, Glass’ delta—which is unaffected by unequal variances—was used as a measures of effect size. Due to the relatively large age range, we also present sex differences in ADI-R and BRIEF scores separately for girls and boys pre- and post-puberty (11 years and younger and 12 years and older) (See Supplementary Tables S8 and S9).

To investigate the association between ADI-R sub-scores (i.e., reciprocal social interaction, communication, and RRB) and EF (BRIEF GEC), we first calculated a Pearson correlation coefficient. To investigate the impact of covariates (i.e., sex, IQ, age, ADHD, and a sex * EF interaction) on the association between ADI-R sub-scores and BRIEF GEC, we fitted a series of nested multiple regression models and then compared the fit of these models by calculating Akaike information criterion (AIC) values and F-ratios for model change. For sex, “male” was the reference category, and for the ADHD variable, “no ADHD diagnosis” was the reference category. Lower AIC values are indicative of better model fit. As we were interested in three sub-scores from the ADI-R for these multiple regression models, we adjusted the critical value for 3 tests (ADI-R A, ADI-R B, ADI-R C) (critical p-value = 0.017), with values less than 0.05 considered on the border of statistical significance for the purposes of these analyses (i.e., Bonferroni correction). Although this is an arbitrary cutoff for values considered to be on the border of statistical significance, we chose 0.05 as this is the value traditionally used when analyses are not corrected for multiple comparisons. To generalise the regression results beyond the given samples, robust regression was performed in the event of non-normally distributed standardized residuals via bootstrapping with 2000 samples. We obtained bootstrapped 95% confidence intervals for the model intercept and slopes and compared these with the confidence intervals from the original model when the standardized residuals from models were not normally distributed. Confidence intervals for the intercept and slopes of these models were similar to a bootstrapped model, indicating that there were no considerable problems with non-normal distribution of residuals in the model (See Supplementary Tables S5–S7 for details). To examine the impact of more closely matched boys and girls on age and IQ, the same model fit and comparison procedure was performed on a subset of the sample, which was generated using the FUZZY extension command in SPSS. These analyses can be found in the supplement section. We allowed cases to be matched on age within 2 years and total IQ within 10 points. Three girls had missing full-scale IQ data, so the 22 girls with no missing values were matched to 44 boys (Supplementary Table S1). Analyses of the BRIEF subdomains behavioral regulation index (BRI) and metacognition index (MI) for the total sample are shown in Supplementary Tables S10–S15.

There were no statistically significant differences between sexes (critical alpha adjusted to p = 0.008) in any of the ADI-R domains, BRIEF GEC, full-scale IQ, or age (Table 1). However, there were tendencies for girls to be slightly older (p = 0.029), have some more difficulties on the BRIEF index MI (p = 0.045) and to have less difficulties with the ADI-R C domain restricted and repetitive behavior (p = 0.038) than the boys, but these sex differences did not reach the adjusted significance level. We also found no significant differences between girls and boys in any of the ADI-R domains, BRIEF index scores, full-scale IQ, or age when we split the sample into pre-puberty and post-puberty groups (Supplementary Tables S8 and S9).

There was no significant diffference in the proportion of boys and girls with comorbid ADHD (χ² = 2.96, p = 0.09).

There was a statistically significant correlation (adjusted critical alpha = 0.017) between reciprocal social interaction and EF (r = 0.31, p < 0.001), as indexed by scores on the ADI-R-A and BRIEF GEC, respectively. We fitted three nested linear regression models to assess the role of covariates (i.e., sex, IQ, age, and ADHD diagnosis) and the interaction of sex and EF on the relationship between reciprocal social interaction and BRIEF GEC (Table 2). The first model, which included sex, IQ, age, and ADHD diagnosis, was not statistically significant (p = 0.49). The second nested model, which added BRIEF GEC, was on the border of our adjusted statistical significance threshold (p = 0.04). The second model (AIC = 630.9) was a significantly better fit of the data than the first model (AIC = 637.4; F(1, 96 = 8.38, p = 0.005), indicating that EF is related to reciprocal social interaction, over and above the main effects of sex, IQ, age, and ADHD diagnosis. The third nested model, which added the interaction of BRIEF GEC and sex, significantly predicted social interaction (p = 0.001). In this model, BRIEF GEC, sex, and their interaction provided a statistically significant contribution (Table 2). The third model (AIC = 619.7), which included a sex * BRIEF GEC interaction term, was a significantly better model for the data than the second model, which only included main effects (AIC = 630.9; F(1, 95) = 13.15, p < 0.001). Simple slopes analysis revealed a statistically significant slope for females (0.47, SE = 0.1, t = 4.63, p < 0.0001) but not for males (0.05, SE = 0.05, t = 0.95, p = 0.35). In other words, a one-point increase in ADI-R-A score is associated with a 0.47 BRIEF GEC score increase in females, but only a 0.05 BRIEF GEC score increase in males.

The same model fit and comparison procedure on subset of participants more closely matched on age and IQ revealed similar results (Supplementary Table S2) (Fig. 1).

There was a statistically significant correlation (adjusted critical alpha = 0.017) between communication and EF (r = 0.33, p < 0.001), as indexed by scores on the ADI-R-B and BRIEF GEC, respectively. We fitted three nested linear regression models to assess the role of covariates and the interaction of sex and EF on the relationship between ADI-R B and BRIEF GEC (Table 3). The first model, which including sex, IQ, age, and ADHD diagnosis, was not statistically significant (p = 0.84). Although the second nested model was also not statistically significant (p = 0.20), BRIEF GEC provided a contribution that was on the border of statistical significance (p = 0.02). This second model (AIC = 577.3) was a better fit of the data than the first model (AIC = 581.5; F(1, 92) = 5.98, p = 0.02), indicating that EF is related to communication, over and above the main effects of sex, IQ, and ADHD diagnosis. However, this effect was on the border of statistical significance (p = 0.02) and needs to be validated in future studies. The third nested model, which added the interaction of BRIEF GEC and sex, significantly predicted communication (p = 0.004). In this model, BRIEF GEC, sex, and their interaction provided a statistically significant contribution (Table 3). The third model (AIC = 566.9), which included a sex * BRIEF GEC interaction term, was a significantly better model for the data than the second model, which only included main effects (AIC = 577.3; F(1, 91) = 12.27, p = 0.001). Simple slopes analysis revealed a statistically significant slope for females (0.38, SE = 0.09, t = 4.29, p < 0.0001) but not for males (0.03, SE = 0.05, t = 0.61, p = 0.55). In other words, a one-point increase in ADI-R-B score is associated with a 0.38 BRIEF GEC score increase in females, but only a 0.03 BRIEF GEC score increase in males. The same model fit and comparison procedure on subset of participants more closely matched on age and IQ revealed similar results (Supplementary Table S3).

The correlation between restricted, repetitive and stereotyped behavior and EF, as indexed by scores on the ADI-R–C and BRIEF GEC respectively, was on the border of the adjusted critical alpha (r = 0.22, p = 0.019; adjusted critical alpha = 0.017). We fitted three nested linear regression models to assess the role of covariates and the interaction of sex and EF on the relationship between repetitive behavior and EF (Table 4). The first model, which including sex, IQ, age, and ADHD diagnosis, was not statistically significant (p = 0.43). Nor was the second nested model which added BRIEF GEC (p = 0.12). This second model (AIC = 439.9) was a better fit of the data than the first model (AIC = 443.1; F(1, 93) = 5.08, p = 0.03), but was on the border of statistical significance. The third nested model (adding the interaction of BRIEF GEC and sex) was not statistically significant (p = 0.06) (Table 4). The third model (AIC = 438.4) was a better fit of the data than the second model (AIC = 439.9), but this was not statistically significant (F(1, 92) = 3.3, p = 0.07). Simple slopes analysis revealed a statistically significant slope for females (0.12, SE = 0.05, t = 2.65, p = 0.01) but not for males (0.03, SE = 0.02, t = 1.15, p = 0.25). In other words, a one-point increase in ADI-R–C score is associated with a 0.12 BRIEF GEC score increase in females, but only a 0.03 BRIEF GEC score increase in males. The same model fit and comparison procedure on subset of participants more closely matched on age and IQ revealed similar results (Supplementary Table S4).

The finding that executive dysfunction and social difficulties are highly related in girls but not in boys might be important for various aspects of clinical practice. Firstly, when girls present with high scores on the ADI-R, it is reasonable to assess for executive difficulties and vice versa. Furthermore, because girls might have a higher risk for executive dysfunction in combination with their social difficulties, the finding can have implications for the choice of interventions. Following this argument, it is possible that girls (with the same degree of social difficulties as boys) will benefit more from EF interventions. Some existing programs that aim to enhance EF have shown to be effective on both social problems and EF (Kenworthy et al., 2014). However, to our knowledge, research is yet to investigate whether this treatment may be more effective for girls than boys. Future studies need to consider that sex differences might influence the effect of interventions.

The study recruited a clinically well-defined sample of children and adolescents with ASD, however, there were still relatively few girls included in the study. The participants were recruited from specialist health care services, which may limit the results to more severe conditions as previous studies have shown that girls referred to specialist clinics have more severe problems than boys (Wang et al., 2017). The girls in our study were slightly older than the boys, but age was accounted for in the nested linear models. The BRIEF is based on parent’s own observations and evaluations of the child. This parental bias might have influenced the findings, but on the other hand, this instrument has been shown to be an ecologically valid measurement of how the child functions in everyday life. We have used the t-score from the BRIEF in the analyses, which have age and gender “corrected” norms, since t-scores are commonly used in literature, as well as clinical practice, and it is important to understand how different clinical tools influences each other. As our study included a relatively small sample of females and a large age range, we also performed a FUZZY match analysis to investigate the phenomena between the females and males in a more closely matched sample on age and IQ. Similar findings from the FUZZY match sample and the total sample contributes to the validity of our results, even though we had a relatively small sample of females and a large age range. Both the BRIEF and the ADI-R are based on information from parents and this might bias the findings. However, while the BRIEF is a questionnaire, the ADI-R is a clinical semi-structured interview, which involves a clinical rating. Together, they both give important information about a child’s behavior. In future studies it would be interesting to investigate possible sex differences in the relationship between the semi structured observation ADOS, which is a direct measurement of the child’s behavior, and BRIEF scores. In this way, one can further investigate if our findings are related to actual sex differences or whether it is more an expression of different parental reporting based on sex. Unfortunately, this was not possible in the current study. We did not have access to ADOS item-level scores and therefore were not able to calculate comparable severity scores across the different ADOS versions and modules used in this study.

Another potential reason for the sex difference in ASD prevalence might be that girls have a different phenotype. Currently, the established diagnostic practices and tools like the ADOS and the ADI-R are not constructed or adapted to measure the subtle difficulties that girls may present with, which differ from the typical presentation of ASD symptoms in boys. Lai and colleagues suggest this might be a circular phenomenon, since an ASD diagnosis is based on behavioral descriptions, and the most common diagnostic tools are largely validated on the classic male phenotype of autism behaviors (Lai et al., 2015). Since we did not include TD children and adolescents in our study, we do not know if our findings are specifically related to ASD or if there is a more general sex difference between social dysfunction and EF which applies to TD children as well as other diagnostic groups.