Executive Function: Cognition and Behaviour in Adults with Autism Spectrum Disorders (ASD)

Participants with ASD were recruited from a specialist diagnostic service for adults. This was a clinical sample of consecutive participants meeting inclusion criteria over a period of 3 years, all of whom completed the same neuropsychological and diagnostic battery and consented to take part in the study All participants received a clinical diagnosis of ASD made by a consultant psychiatrist and nurse specialist and met ICD-10 criteria for Autism (Autism; n = 26), Asperger’s syndrome (AS; n = 53), atypical autism (n = 21) or pervasive developmental disorder not otherwise specified (PDD-NOS; n = 10). Where a reliable informant-based developmental history was available the autism diagnostic interview-revised (ADI-R; Lord et al. 1994) was completed (n = 78) and in cases where this was not available then participants underwent the autism diagnostic observation schedule (ADOS; Lord et al. 1989; n = 48). In a small number of cases for clinical soundness, both the ADI-R and the ADOS were completed (n = 19) with both measures showing agreement. In three cases where one or both measures were sub-threshold on one domain, diagnoses of PDD-NOS, Asperger’s syndrome and atypical autism were given. All participants received their diagnosis in adulthood. Mean ASD group scores on the ADI-R were as follows [ADI-R Communication Domain Mean Score 10.08 (SD = 5.11); ADI-R reciprocal social interaction domain mean score = 12.47 (SD = 6.01); ADI-R repetitive behaviour domain mean score = 3.58 (SD = 2.03)]. Of the ASD group completing the ADI-R, 25 (32%) of individuals scored below diagnostic threshold on the communication domain, 22 (28%) scored below threshold on the RSI domain and 25 (32%) on the RB domain. Where ADI-R scores were below diagnostic threshold, diagnosis was supplemented with the ADOS-G (n = 5) with above threshold scores for autism spectrum disorder. Twenty-one (19%) of the sample received a diagnosis of atypical autism in keeping with below threshold scores on one domain of the ADI-R. Five individuals scored below diagnostic threshold on > 1 domain on the ADI-R and ADOS-G scores were not available. Of those completing the ADOS-G (n = 48, 43.6%), mean ADOS communication and social interaction summary score = 11.10 (SD = 4.09).

Psychiatric co-morbidity was assessed via clinical interview and recorded as ICD-10 diagnosis on the case file with the exception of attention deficit hyperactivity disorder (ADHD), which was made according to DSM-IV criteria as per National Institute for Health and Care Excellence (NICE) guidance.

Exclusion criteria were: Verbal IQ < 80 (given the lack of normative data for EF measures for people with low IQ), any other neurodevelopmental disorder diagnosis such as attention deficit hyperactivity disorder (ADHD), Fragile X syndrome and/or Velocardio facial syndrome, any prior brain injury, diagnosis of psychosis, substance misuse, and/or an eating disorder.

The Control group was recruited from a volunteer database of individuals willing to participate in research, and an opportunity sample. The two groups were matched by age, gender and verbal IQ (see Table 1). The study had ethical approval from the local ethics committee. Participants were provided with information about the study and asked to provide written consent to participate if they agreed.

There was no formal sample size calculation. However, as we expected relatively high communality across the sub-domains of EF, a sample size of 100 was considered appropriate as per guidance (MacCallum et al. 1999).

The following measures were used to assess intellectual ability and executive functioning. All measures were part of a clinical neuropsychological test battery used as part of a wider autism diagnostic assessment and have been previously used in studies with autistic adults. Tests were chosen to have high ecological validity given that this should allow for the best association with everyday living skills.

All participants were tested in a quiet room by an assistant psychologist trained in test administration and working under the supervision of a qualified clinical psychologist.

Data were initially analysed to examine between group differences in mean scores on the main variables. The proportion of individuals meeting criteria for clinical abnormality on each domain of executive function was also compared across groups. To investigate the hypothesised variance in performance across executive function tests, exploratory factor analysis was conducted on the summary test scores of the main 5 EF measures. A single component was extracted which allowed a single variable ‘EF function’ to be created. This variable was entered into regression equations to examine its contribution to the variance in self-report behavioural measures of executive difficulties and scores on autism measures.

Table 2 shows the median and interquartile range for each group across the test battery. Mann–Whitney tests were used to compare group differences due to non-normal data distribution on the main EF measures. T-tests were used to consider group differences on variables that were normally distributed. Table 2 shows that there were significant between group differences across a range of EF tasks. Large group differences were particularly seen on measures of verbal generativity (FAS), verbal initiation and inhibition (Hayling) and cognitive flexibility (Brixton).

Scores below the 5th percentile in the majority of neuropsychological tasks are generally taken as indicative of clinically meaningful impairment and have been used in similar studies (e.g. Hill and Bird 2006). Figure 1 shows the percentage of people falling within the clinically impaired range on each of the EF tasks according to published test norms. Chi squared analysis was used to compare groups; significantly greater numbers of participants in the ASD group scored in the clinically impaired range on the FAS χ _{(1, N = 141)} ² = 11.86 (p < 0.001) and Hayling 1 χ _{(2, N = 141)} ² = 9.41 (p < 0.01) tests. The Hayling 2 χ _{(2, N = 141)} ² = 4.73 (p < 0.05), Hayling Total χ _{(2, N = 141)} ² = 7.35 (p < 0.05), Zoo Map χ _{(1, N = 141)} ² = 5.96 (p < 0.05) and Brixton χ _{(2, N = 141)} ² = 7.35 (p < 0.05) tests were significant but not at reduced α (0.01) to correct for multiple comparisons.

To address the possibility that impairment might be heavily loaded on a few individuals within the ASD group, analyses were conducted to examine whether the groups differed in distribution of clinical impairment (see Table 3) (χ _{(5, N = 141)} ² = 24.32, p < .0001). Of note, more than one-third of the ASD group were not in the clinically impaired range on any of the EF tasks.

In order to examine possible timing differences on the measures of executive function between the two groups, a ‘time’ variable was created by adding the time in seconds taken by each participant to complete all timed tasks. The mean ‘time’ score for participants with ASD differed significantly from that of the Control group, ASD mean task time = 500.6 s (range 119–1518 s) and Control group mean task time = 365.3 s (range 166–946 s); t_(75.4) = − 3.64, p < .001, 95% confidence interval of the mean difference lower = − 215.7, upper = − 55.27.

Table 4 shows the results of self- and informant-report DEX questionnaires. On both self and informant versions, the ASD group was reported to have significantly greater difficulties on all sub-scales of the DEX measure. There is a large amount of missing data for the DEX due to administrative issues unrelated to participants.

The standardisation data for the DEX suggested that the informant scale can be taken as a more reliable indicator of difficulties in individuals with reduced insight. Paired samples t-tests suggested that a significant difference between self and informant ratings was evident in both the ASD and the Control group albeit in opposing directions; ASD group mean self-informant discrepancy = − 4.99 (SD = 13.65), t = − 2.45, df = 43, p < .05; Control group mean self-informant discrepancy = 6.44 (SD = 9.001), t = 3.13, p < .01.

Normative data for self-report for individuals with brain injury on the DEX is 27.21 (SD = 14.48) and informant-rated 32.85 (SD = 15.98). In the present study, the group with ASD and their informants report levels of characteristics associated with dysexecutive syndrome similar to those reported by people with acquired brain injury.

Further exploration of participants who had completed the DEX questionnaire with co-morbid psychiatric diagnosis information available in the clinic report revealed no significant differences between the mean scores for the ASD group with a co-morbid psychiatric diagnosis (n = 30) and those without a co-morbid diagnosis (n = 31) (ASD co-morbid group mean DEX total self-rating score = 30.37 (SD = 11.15); ASD non co-morbid group mean DEX total self-rating score = 27.22 (SD = 11.70); t = − 1.076, df = 59, p = .286). The same non-significant finding was also present when DEX informant ratings were compared between the co-morbid and non-co-morbid groups.

All 5 main EF measures (Zoo Map and Key Search total profile scores, total number of correct words on the FAS, Hayling overall total score and Brixton scaled score) were significantly correlated with each other (see Table 5).

Exploratory factor analysis using principal components analysis was performed to consider factors underlying performance on the 5 domains of EF function for all participants. Initial Eigen values indicated that the 1st component explained 46% of the variance. The remaining 4 components had Eigen values < 1.0 and thus just 1 factor was extracted from the data and although equamax rotation was specified, the single factor solution could not be rotated. This single factor was used to produce a regression factor score, creating a variable labelled ‘EF function’. The single factor underlying the 5 EF tests is consistent with Baddeley’s (1974) construct of a single overarching executive function.

The ASD and Control groups differed significantly in terms of mean scores on ‘EF function’ (see Fig. 2). (ASD mean EF score = − .20734 (SD = 0.993), Control group mean EF score = 0.72235 (SD = 0.114); (equal variances not assumed Levine’s f = 6.758, p = .01); t = 6.331, df = 78.457, p < .0001, 95% CI of the difference 0.6373–1.222).

To further consider the role ASD symptoms might contribute to executive difficulties, ADI-R scores, number of co-occurring psychiatric conditions, age, Verbal IQ and the ‘time’ variable were entered into a regression equation. The model was significant (F = 27.12, p < .0001, R² = .739) with Verbal IQ being a positive predictor of EF and time being a negative predictor of EF. None of the other variables made a significant contribution.

As already noted, a clinical presentation characteristic of executive function difficulties as measured by the DEX rating scales was prevalent in the ASD group. Considering the variables that might contribute to the variance in the behavioural, cognitive and emotional features rated on the DEX, multiple linear regression was used. The ‘EF function’ score, time score, ADI communication, reciprocal social interaction and repetitive behaviour scores were entered into the equation. The model was non-significant F = .971, p = .458; with R² = .139 suggesting that neither performance on neuropsychological tests of executive function, speed of processing nor ASD symptoms contributed to the DEX self-report total rating score. A non-significant finding was similarly found when the dependent variable was specified as the DEX informant report total score.

In relation to our broad research question, investigating the presence and pattern of EF performance, across multiple domains of EF measurement we found that adults with ASD had lower scores relative to matched controls across measures of planning (Zoo Map, Key Search), generativity (Hayling test, verbal fluency) and flexibility (Brixton). A significantly greater proportion of the ASD group than controls had scores in the ‘clinically abnormal’ range on a measure of generativity (35% impaired) and a measure of response initiation (24%). These measures require generation of a verbal response, with the former less scaffolded than the latter. Clinical impairment was widespread in the ASD group, 20–30% had clinically impaired scores on measures of planning, 20% on a measure of flexibility with fewer impaired on a measure of verbal inhibition (15%). The results of a factor analysis across all participants indicated a single factor underlying performance on the neuropsychological domains measured in this study and the ASD group differed significantly from the Control group on this factor.

In support of hypothesis 1, adults with ASD took significantly longer to complete EF tasks. Since both groups were matched for IQ and age this slow performance style appears to be consistent with previous reports (e.g. Johnston et al. 2011; Hill and Bird 2006) noting a slow and accurate response style amongst individuals with ASD.

Individuals with ASD reported high levels of dysexecutive difficulties on the DEX questionnaire. However, scores on the behavioural measure of dysexecutive syndrome did not bear close relation to performance on tests of executive function, speed of task performance or to measures of ASD symptoms and thus our hypothesis was not supported.

It is important to note that one-third of the group with ASD were not in the clinically impaired range on any of the EF tasks which is at odds with other similar studies (e.g. Hill and Bird 2006). This may reflect differences in the way that clinical impairment was calculated in our study (using normative data from test manuals) and a larger sample size. The exclusion of participants with co-occurring ADHD may also be relevant to this finding, particularly the relatively low rates of impairment on a test of inhibition.

In order to exclude the possibility that these results are explained by a small number of individuals performing poorly on EF tasks we examined the number of individuals who had scores in the clinical impairment range across 0, 1, 2, 3, 4 or all 5 EF measures. Results showed that whilst none of the control sample were impaired on more than one measure, a third of participants with ASD were impaired at two or more tests. Interestingly very few individuals with ASD were impaired across all measures suggesting variance within executive functioning as might be expected given such a heterogeneous range of tasks assessed and the variability within the autism spectrum. This is consistent with the results of a recent meta-analysis (Demetriou et al. 2018) where combining findings from a wide range of studies highlighted a broad spectrum of EF impairment in ASD.

Consistent with our hypothesis that adults with ASD would show slow task completion is the finding that the most common EF impairment in the ASD group was on a task of verbal fluency (FAS) which is considered a measure of processing speed as well as a measure of generativity (Spek et al. 2009). Score attainment on this verbal fluency task is time-sensitive. There was also a significant difference between groups on the time taken to complete the Zoo Map 2 task. We calculated an overall timing score on the basis of timed, open-ended tasks i.e. where time taken did not influence performance scores. Taken together, these findings suggest that on both open-ended and time-constrained neuropsychological measures, the ASD group were significantly slower. Our ‘Time’ variable was significantly negatively associated with the overall ‘EF’ factor, suggesting that slow and effortful processing was not supportive of more accurate task performance.

In relation to dysexecutive difficulties, informants rated these as being somewhat higher than the individuals themselves in the ASD group. This is consistent with normative data for the measure. However the converse discrepancy between self- and informant-ratings was present in the Control group and thus it is hard to draw conclusions. Self- and informant-rated scores in the ASD group on the behavioural measure of dysexecutive difficulties were very similar to those reported by individuals with acquired brain injury. In the brain injury literature, performance on more traditional tests of neuropsychological function have not always been found to align with functional difficulties (Shallice and Burgess 1991a, b). We used more ecologically valid measures of EF in the present study to try and overcome this methodological issue. Previous studies with children have shown a strong association between the BRIEF and core ASD symptoms (Gilotty et al. 2002) and also failed to find an association between everyday executive function and ASD symptoms in children (van den Bergh et al. 2014). Our findings are consistent with those of Wilson et al. (2014) who reported a similar lack of association between impaired performance on cognitive assessments of EF and ASD symptoms in adults. Curiously, the timing variable while relevant to EF task performance did not account for variance in dysexecutive behaviour scores. Furthermore, our findings do not appear to reflect a lack of ‘insight’ into dysexecutive problems on the part of autistic adults as the informant-rated DEX scores also showed no association with ASD symptoms. Further investigation of the items on the DEX questionnaire is warranted to elicit a greater understanding of the phenomenology and potential underpinnings of high scores by adults with ASD. Although not directly related to ASD symptoms as measured in this study, the items potentially are not related to neurocognitive indices either. Whilst these are clinically interesting findings, an important caveat is the large amount of missing data for the DEX this is in part due to administrative processes rather than issues related to participants. These findings should therefore be interpreted with caution as this is not necessarily a representative sample of adults with ASD.

Our findings showed that individuals in the ASD group were impaired across measures with both implicit and explicit expectations. For instance, 20% of our sample fell in the clinical abnormality range on the Brixton task which has clear instructions that are much more explicit than other measures of cognitive flexibility such as the widely used Wisconsin Card Sorting Test (WCST). Similarly, the Zoo Map test is considered a more ‘constrained’ measure of planning and is less often impaired in individuals with ASD (White 2013), but our findings indicated that it was on this test that the ASD group showed the second highest level of impairment. On Zoo Map 2 which is a more structured task, participants with ASD made relatively more errors and performed more slowly than control participants although these differences did not reach significance. Interestingly the more open-ended task (Key Search) was the only one where participants with ASD did not show statistically significantly different rates of clinical abnormality as compared to age and IQ matched controls.

The findings of the present study and conclusions that can be drawn are limited by the relatively narrow and blunt measure of executive functions utilised in our clinical sample. We used well validated and standardised clinical neuropsychological measures of executive function to investigate our hypotheses. Although this brings advantages in respect of ecological validity, the complex and inter-related nature of executive processes means precise conclusions relating to cognitive models of EF cannot be drawn from these broad measures. Moreover, the clinic diagnostic algorithm meant ADI-R and ADOS scores were not available for all participants uniformly, and these analyses are limited in this respect. Relatedly, the use of older assessment tools (ADOS-G and WAIS-III) in the current study reflects the length of time over which this data was collected. Studies comparing these tools with their current counterparts have generally shown high rates of agreement (e.g. correlation coefficients ranging from .83 to .94 for WAIS-III and WAIS-IV; Wechsler 2008) or there have been no significant changes (e.g. ADOS-G to ADOS-2, module 4) meaning that the current findings are relevant to current criteria/practice. Forthcoming changes to the classification of autism in the ICD-11 will be relevant in understanding the nature and pattern of both executive functioning and everyday living skills in autism given the discontinuity in diagnostic criteria. It is suggested that a significant proportion of individuals who currently meet diagnostic thresholds will no longer do so under ICD-11 (or DSM-5) criteria (e.g. Doernberg and Hollander 2016; Wilson et al. 2013). This raises important issues about the generalisability of existing research in this area and access to clinical/educational services.

The results of the current study do bring some clarity to the literature on EF function in ASD. In studying adults without intellectual disability and with robust diagnostic assessment of both ASD and relevant co-morbidity, particularly excluding ADHD, important and relevant issues are highlighted for people with ASD and clinicians. Individuals with ASD reported high levels of dysexecutive symptoms that were functionally impairing in everyday life across behavioural, cognitive and emotional domains. A considerable proportion of the adults with ASD in this study were impaired across one or more neuropsychological measures of executive function. Generativity and response initiation were most affected in respect of clinical impairment. These data present preliminary evidence that performance on EF measures appears to reflect a single underlying construct. Functional and neurocognitive indices of EF impairment were independent of ASD symptoms and of each other. These findings suggest that EF is an additional important co-occurring condition to consider in ASD. Of equal importance is the finding that one-third of adults were not in the impaired range on any of the neurocognitive measures. Future research seeking to more precisely understand the nature of EF in ASD should seek to delineate assessment paradigms further by reducing the influences of processing speed, theory of mind ability and implicit task demands on task performance. It would also be useful to explore differences in executive functioning between adults with both ASD and ADHD, ASD only and controls to further understand the pattern of EF strengths and difficulties and the relative contribution made by each neurodevelopmental condition. Future investigation into the nature of everyday ‘dysexecutive’ behaviours is warranted. At a clinical level, it would seem beneficial to ensure that EF assessment is multi-method and that impairments are conceptualised as (a) potentially significant in the formulation of clinical problems and (b) important to scaffold when modifying psychosocial interventions to meet the needs of autistic people.