Abstract
Autism Spectrum Disorders (ASD) and ADHD are highly heritable developmental disorders that frequently co-occur (Ames & White, 2011; Ronald, Simonoff, Kuntsi, Asherson, & Plomin, 2008). Twin studies reveal a moderate degree of phenotypic overlap between ASD and ADHD both throughout the whole range of scores and at the upper extreme end (Reiersen, Constantino, Grimmer, Martin, & Todd, 2008; Ronald et al., 2008), and there is evidence for shared etiological factors for ASD and ADHD (Rommelse, Franke, Geurts, Hartman, & Buitelaar, 2010; Rommelse, Geurts, Franke, Buitelaar, & Hartman, 2011; St. Pourcain et al., 2011). Furthermore, clinical studies have documented poor social skills, language delay, sensory overresponsivity, attention problems, oppositional defiant behavior, and emotion regulation problems in both ASD and ADHD (Gadow, DeVincent, & Schneider, 2009; Mulligan et al., 2009; Rommelse et al., 2011). An increasing number of studies showed an overlap between ASD and ADHD with respect to cognitive functions (Booth & Happé 2010; Corbett, Constantine, Hendren, Rocke, & Ozonoff, 2009; Fine, Semrud-Clikeman, Butcher, & Walkowiak, 2008), and therefore, studying ASD and ADHD together may provide the most optimal strategy in examining both shared and unique underpinnings. For an in-depth discussion on differing models of co-existence of ASD and ADHD, see also Banaschewski, Neale, Rothenberger, and Roessner (2007) and Rommelse et al. (2011).
ASD and ADHD are both highly heterogeneous disorders; however, the optimal approach to describe this heterogeneity remains unclear. Latent class analyses (LCA) have been used with the aim to identify more homogeneous subgroups of both traits. This approach in the separate fields of ASD and ADHD research previously resulted in the identification of subgroups that mainly differ by disorder severity rather than truly distinct categories in a general population sample (Acosta et al., 2008; Constantino, 2011; Volk, Todorov, Hay, & Todd, 2009). Recent studies using both ASD and ADHD symptom measures also disclosed concordant classes differing mainly in severity (Mulligan et al., 2009; Reiersen, Constantino, Volk, & Todd, 2007). Such concordant ASD–ADHD trait profiles highlight the shared etiology of both traits, with both disorders sharing a common genetic and biological basis. In contrast, discordant ASD–ADHD trait profiles, that are highly symptomatic on one trait but not the other, may have atypical underpinnings. These underpinnings may translate into differential prognoses and susceptibility toward treatment. Some successes have been made in the identification of such discordant subtypes using LCA or LCGA (Latent Class Growth Analyses), adopting a longitudinal perspective (St. Pourcain et al., 2011; van der Meer et al., 2012). St. Pourcain and colleagues (2011) suggested that children with ADHD symptoms without ASD symptoms may more often have a childhood-limited form of ADHD, while children having high ASD and ADHD symptoms may have more persistent ADHD symptoms, and possibly are more resistant toward treatment. Furthermore, we previously used both ASD and ADHD clinical symptom measures as well as measures for comorbid internalizing and externalizing problems in LCA and identified among others two mutually exclusive classes of children with clinical symptoms of both ASD and ADHD. One of these had proportionally more ASD than ADHD symptoms and the other just the other way round (van der Meer et al., 2012). Importantly, these classes showed opposite visual-spatial processing capacities, suggesting the identification of behavioral subtypes may increase our understanding of the cognitive heterogeneity in both disorders as well as the etiology of their co-occurrence.
A frequently overlooked issue that may have hindered progress in identifying more homogeneous, etiologically distinct disorder subgroups is that not only ADHD and ASD populations, but also unaffected populations are characterized by heterogeneity. Unaffected populations are usually described with lack of precision and usually lumped together into a single group without symptoms. This hinders the study of heterogeneity in this group, and ignores strong evidence that ASD and ADHD as well as other internalizing and externalizing behavioral disorders exist on a continuum (Constantino, 2011; Levy, Hay, McStephen, Wood, & Waldman, 1997; Lundstrom et al., 2012; Plomin, Haworth, & Davis, 2009; Robinson et al., 2011). Hence, cognitive and symptom heterogeneity at the upper end of the symptom distribution (i.e., in the clinical range) may well be reflective of similar cognitive and symptom heterogeneity at the lower end of the distribution. Such cognitive heterogeneity was recently examined in a sample of ADHD and typically developing children (Fair, Bathula, Nikolas, & Nigg, 2012). Individual-based analyses on a range of cognitive tasks revealed that some of the cognitive heterogeneity in children with ADHD seemed to be nested within the variation in typically developing children: Largely similar cognitive subtypes (i.e., neuropsychological subgroups) were revealed in both populations. The authors also showed that diagnostic accuracy increased somewhat when the ADHD versus control contrast was made within each cognitive subtype instead of whole group analyses. Furthermore, the study showed that a large part of the previously unexplained cognitive heterogeneity within ADHD seems actually not to be related to ADHD as a disorder (i.e., the upper end of the symptom distribution), but more so to cognitive heterogeneity also present in the non-clinical part of the ADHD spectrum.
Ordinary assessments of psychopathology would disguise such heterogeneity across the continuous ASD and ADHD traits, as they give resolution only to the affected end of disorders, reflected in skewed distributions of symptom measures. An exception are questionnaires that provide greater resolution across the entire distribution, taking into account difficulties as well as possible strengths such as well-developed social-communication and attention traits. This approach seemed quite promising in distinguishing ADHD subtypes in population data on the Strengths and Weaknesses of ADHD Symptoms and Normal Behavior (SWAN) rating scale (Arcos-Burgos et al., 2010), which revealed latent classes with less than average hyperactivity and impulsivity. While the former study focused on symptom data, further work on identifying distinct non-affected subtypes may also provide us with a better understanding of the previously unexplained cognitive heterogeneity in ASD and ADHD. Therefore, the present study aimed to identify distinct ASD–ADHD trait profiles that are typified by distinct cognitive and behavioral profiles across the ASD and ADHD trait continua, including the unaffected ends. These children were also part of our previous study in a combined clinical and population-based sample, where more than 80% of them were lumped into “normal classes” on the basis of clinical ADHD and ASD measures (van der Meer et al., 2012).
In the present study, our focus is on the population-based sample, studying it using measures that are sensitive assessments across the continuous ASD and ADHD traits. LCA were used to identify distinct ASD–ADHD profiles in the general population, and these profiles were examined for their internalizing and externalizing problem and cognitive correlates. Given the correlations usually reported between both quantitative traits and the comorbidity between ASD and ADHD as extreme ends, it was expected that mostly concordant ASD−ADHD classes would be detected that differed quantitatively but not qualitatively. However, by providing greater resolution of scores across the trait continua, we also expected to have greater power to identify discordant classes with distinct behavioral and cognitive profiles.
Method
Participants
The study has been approved by the Central Committee on Research involving Human Subjects (CCMO) and participants were enrolled between January 2009 and July 2011. Eligible participants were 378 children from a random population cohort study (Schoolkids Project Interrelating DNA and Endophenotype Research [SPIDER]). All children were between 6 and 13 years of age (Mage = 8.9 [1.7], % male = 49.5). All were of Caucasian descent and had an estimated total IQ of at least 70 on the Wechsler Intelligence Scale (WISC-III; Wechsler, 2002). Exclusion criteria were epilepsy, known genetic or chromosomal disorders, brain damage, and problems with vision or hearing. After complete description of the study to the parents, written informed consent from all parents was obtained.
Measures
ASD and ADHD symptom measures
ASD and ADHD trait measures according to parents were obtained using the Autism Quotient (AQ; Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley, 2001; Hoekstra, Bartels, Cath, & Boomsma, 2008) which provides a quantitative measure of ASD-like traits in the general population, and the SWAN rating scale (Hay, Bennett, Levy, Sergeant, & Swanson, 2007), respectively. Both measures have shown adequate reliability and validity (Arnett et al., 2011; Hoekstra et al., 2008). Both are based on a Likert-type rating scale and show scores that followed a continuous distribution in the general population (Baron-Cohen et al., 2001; Polderman et al., 2007). The distribution of these measures resembled a Poisson distribution rather than a normal distribution. Therefore, we modeled the subscales as count variables in the latent class model (Muthén & Muthén, 2006).
Cognitive measures
The six neurocognitive tasks analyzed in this study have been described elsewhere (van der Meer et al., 2012), a brief description of all cognitive dependent variables is provided in Table 1. Ceiling effects did not occur on any of the tasks as indicated by boxplot analyses on raw data.
Description of the Cognitive Measures.
WISC-III = Wechsler Intelligence Scale, Wechsler (2002).
Other internalizing and externalizing problems
In addition to the normally distributed ASD and ADHD trait measures, two questionnaires measuring clinical symptom levels of ASD, ADHD, Oppositional Behavior, Emotional Lability, Anxiety, Perfectionism, and Psychosomatic Complaints were obtained. These were the Social Communication Questionnaire (SCQ, Lifetime version; parent ratings) and the Conners’ Parent Rating Scale-Revised: Long version (CPRS-R:L), both validated instruments for screening developmental problems (Conners, Sitarenios, Parker, & Epstein, 1998; Rutter et al., 2003).
Procedure
The tasks described were part of the neuropsychological assessment battery used in the SPIDER project. Children completed the battery in approximately 2 hr and the order of task administration was counterbalanced. Due to time constraints, not all tasks were administered to all children. Full-Scale IQ was prorated by four subtests of the WISC-III; Block Design, Picture Completion, Similarities, and Arithmetic. These subtests are known to correlate between .90 and .95 with Full-Scale IQ (Groth-Marnat, 1997; Kaufman, Balgopal, Kaufman, & McLean, 1994). Parents were invited to fill in the aforementioned questionnaires concerning their child’s behavior.
Data Analyses
To identify homogeneous ASD–ADHD trait classes, LCA were used on the raw subscale outcomes of the AQ, ranging from 0 to 30 (Social Skills, Attention Switching, Local Details, Communication, Imagination) and raw subscale outcomes of the SWAN (Inattention and Hyperactive-Impulsive). Subscale scores on the SWAN, ranging from 9 to 63, were mirrored so that the scores on all subscales would imply the same: a higher score was indicative of more symptoms. The LCA were carried out using Mplus version 6.11 (Muthén & Muthén, 2006). Both the probability for a child to belong to each of the classes and the conditional probabilities for children in a particular class to show specific behavior were estimated. Next, children were admitted to the class with the highest probability. Mean subscale sum scores on the seven aforementioned subscales were computed, and presented in a line chart, so that quantitative differences between classes could be examined (Figure 1). Size and significance of class differences on these subscales were determined with a MANOVA.
Class scores on AQ (left) and SWAN (right) subscales.
Next, class differences with respect to age and sex were analyzed to check for possible confounders. The identified classes were examined regarding their cognitive profiles and their internalizing and externalizing problems separately, using ANCOVAs with class-membership as a fixed factor, and age and sex as covariates. IQ was not implemented as a covariate as IQ is inherently confounded with symptoms of ASD and ADHD, and could therefore not be separated from the effect of class (Dennis et al., 2009). Class by age, class by sex, age by sex, and class by age by sex interaction effects were examined and reported if significant. If non-significant, interactions were dropped from the model. Dependent variables were speed and/or accuracy measures for each task separately, or subscale scores on the internalizing and externalizing problems. All dependent variables were successfully normalized and standardized into z-scores by applying a Van der Waerden transformation (SPSS version 20). Some of the outcome measures were mirrored, so that the scores of all variables would imply the same: A higher z-score was indicative of a better performance. Correction for multiple comparisons was applied according to the False Discovery Rate (FDR) controlling procedure with a p value setting of .05 (Benjamini & Hochberg, 1995). Effect sizes were defined in terms of percentage of variance explained (
Results
Identifying Homogeneous Symptom Classes
The LCA on the AQ and SWAN subscales were based on fit and accuracy measures (Nylund, Asparouhov, & Muthén, 2007), and revealed a solution with five classes. Five classes had the best fitting Bayesian Information Criterion (BIC) and sample size adjusted (SSA) BIC values, and entropy (.887), and a bootstrapped Lo-Mendell-Rubin likelihood ratio test p value < .001 (see also Table 2). This, combined with the most informative class profiles and all correlation matrix probabilities > .900, indicated accurate classification. The AQ and SWAN profiles of the classes are presented in Figure 1. For the purpose of simplicity, the classes were labeled. Three concordant classes emerged which had either low, medium, or high levels of both ASD and ADHD traits (see also Table 3). We refer to those as “LL” (Low ASD, Low ADHD; 10.1%), “MM” (Medium ASD, Medium ADHD; 54.2%), and “HH” (High ASD, High ADHD; 13.2%), and two discordant classes with either higher scores on the ADHD traits than on the ASD trait “ADHD > ASD”(18.3%), or higher scores on the ASD trait than on the ADHD trait “ASD > ADHD” (4.2%). The LL-class scored low on both the AQ and the SWAN, the MM-class scored intermediate on both measures, and the HH-class scored relatively high on the AQ as well as the SWAN. Roughly 30% of the children in the HH-class passed the clinical cut-off for both measures. Still, all three classes scored below the clinical cut-off on the ASD and ADHD clinical symptom scales. The ADHD > ASD class scored intermediate on the SWAN and low on the AQ, while the ASD > ADHD class scored relatively high on the AQ and low on the SWAN. Roughly 30% of the children in the ASD > ADHD class passed the clinical cut-off for the ASD measure. Again, all scores were below clinical cut-off on the ASD and ADHD clinical symptom scales, indicating that these distinctions would have been disguised in the ordinary assessments of ASD and ADHD.
Results of Latent Class Analyses on AQ and SWAN Measures.
Note. Entropy refers to classification accuracy; BIC refers to Bayesian Information Criterion; SSA BIC refers to sample size adjusted BIC; VLMR LRT refers to the Vuong-Lo-Mendell-Rubin likelihood ratio test, LMR adj. LRT refers to the bootstrapped Lo-Mendell-Rubin likelihood ratio test.
From a 6 classes solution onwards, the p value may not be trustworthy due to local maxima.
Demographic Characteristics of the Children in the Distinct Classes.
LL is the class with low levels of Autism Spectrum Disorders (ASD) and ADHD symptoms; MM refers to the class with intermediate levels of ASD and ADHD; and HH is the class with relatively high levels of ASD and ADHD symptoms. ADHD > ASD is the class with intermediate levels of ADHD symptoms and low levels of ASD symptoms; ASD > ADHD refers to the class with relatively high levels of ASD symptoms and low levels of ADHD symptoms.
Full-scale IQ was estimated by four subtests of the WISC-III (Wechsler Intelligence Scale; Wechsler, 2002): Block Design, Picture Completion, Similarities, and Arithmetic. These subtests are known to correlate .90 to .95 with Full-scale IQ.(Groth-Marnat, 1997).
The total score on the AQ (Autism Spectrum Quotient) reflected the total amount of ASD symptoms (subscales Social Skills, Attention Switching, Local Details, Communication, Imagination), the clinical cut-off of the total score on the AQ in children is 76 (Auyeung, Baron-Cohen, Wheelwright, & Allison, 2008).
The total score on the SCQ (Social Communication Questionnaire) reflected the total amount of ASD symptoms.
The mirrored total scores on the SWAN (The Strengths and Weaknesses of ADHD Symptoms and Normal Behavior Scale) reflected the degree of ADHD-related symptoms.
T-scores on the CPRS (Conners’ Parent Rating Scale) subscales reflected the degree of domain-specific symptoms. The official cut-off for clinically relevant symptoms is a T-score above 63.
A MANOVA using class as a fixed factor and the ASD and ADHD subscales as dependent variables revealed that, as expected, the five classes differed overall significantly (p < .001). Next, all classes were pairwise compared on the separate ASD and ADHD subscales. Only 11 out of 70 comparisons did not reach significance. Roughly, the non-significant differences were on ASD measures (on the left side of Figure 1) between either the LL and ADHD > ASD class or between the ASD > ADHD and HH-class, or on ADHD measures (on the right side of Figure 1) between the LL and ASD > ADHD class. The distribution of all children across the distinct classes as well as the sex, age, and IQ distributions are provided in Table 3. Boys were overrepresented in the classes with higher levels of ASD and/or ADHD traits (classes HH and ASD > ADHD), whereas girls were overrepresented in the class with the lowest levels of both traits (class LL). When corrected for the influence of age and sex, no changes in differences between the classes were found.
Cognitive Profiles of the Distinct Classes
To test in which cognitive domains the classes overlapped or differed, separate ANCOVAs were used for each cognitive domain, with age and sex as covariates. The discordant classes differed from the concordant classes and from each other in their block pattern performance, F(4, 376) = 5.61, p < .001,
Differences between the classes on measures of block patterns and visual-spatial working memory.
Other Internalizing and Externalizing Problems of the Distinct Classes
Scores on the clinical questionnaire (CPRS) indicated that all classes represent the unaffected side of the continuum: None of the classes scored in the clinical range on any of the subscales (i.e., oppositional behavior, emotional liability, anxiety, perfectionism, and psychosomatic complaints). The concordant classes with intermediate or relatively high scores on the ASD and ADHD traits also presented elevated scores on the other internalizing and externalizing traits. For the discordant classes, the highest levels of the other internalizing and externalizing problems were present in the ASD > ADHD class. In contrast, the ADHD > ASD class did not present increased scores on the other internalizing and externalizing traits. When corrected for the influence of age, sex, class by age, class by sex, and class by age by sex interaction effects, results did not change. Results are also presented in Table 3.
Discussion
The present study examined whether differentiated ASD–ADHD latent classes typified by distinct cognitive and behavioral profiles can be identified in a population-based sample. As hypothesized, the individual-based analyses revealed mostly quantitatively differing, concordant ASD–ADHD classes (77.5%) with either low, medium, or high scores on both traits, and two discordant classes with either more ADHD symptoms than ASD symptoms (18.3%), or more ASD symptoms than ADHD symptoms (4.2%). When comparing the latter two discordant classes, the specific combination of ASD > ADHD was characterized by a superior visual-spatial processing, whereas the ADHD > ASD combination was characterized by inferior visual-spatial processing. Furthermore, the class with elevated scores on both traits (HH) presented a cognitive profile which closely resembled the profile of the ADHD > ASD class. The elevated levels of internalizing and externalizing problems in the classes with either high scores on both traits or more ASD symptoms than ADHD symptoms did not translate into more performance deficits than in the other classes.
Intriguingly, many of the findings in these non-clinical ASD–ADHD classes were an extension of our previous study on clinical ASD–ADHD classes as well as other studies across clinical ASD and ADHD samples (Mulligan et al., 2009; Reiersen et al., 2007; St. Pourcain et al., 2011; Todd et al., 2002; van der Meer et al., 2012). A first parallel was that the discordant ADHD > ASD profile was much more common (18.3%) than the other discordant profile ASD > ADHD (4.3%); the latter profile even remained undisclosed in our previous study. This may suggest that across the general population, just as across the clinical population, most of the children who express ASD-behavior also present the less severe “precursor” of ADHD-behavior, as has been hypothesized previously (Rommelse et al., 2010). A second resembling finding was that the ASD > ADHD class is again characterized by superior visual-spatial functioning, whereas the ADHD > ASD class is characterized by inferior functioning. Thus, superior visual-spatial functioning in children with higher scores on the ASD trait than the ADHD trait holds across the general population and the clinical population alike. This finding is in keeping with recent studies across the general population reporting that a higher level of autistic traits measured with the AQ across the general population was also associated with an enhanced visual working memory (Grinter et al., 2009; Richmond, Thorpe, Berryhill, Klugman, & Olson, 2012). A third similarity was that the class with high scores on both traits showed a cognitive profile that most closely resembled the ADHD > ASD class. This may suggest that children with elevated scores on both traits primarily suffered from cognitive problems with an ADHD-alike etiology. Alternatively, the behavior in the HH-class may actually reflect a true co-occurring of cognitive features of both ASD and ADHD, but the prime cognitive deficits specific for ASD may be obscured by the cognitive deficits most robustly found in ADHD. A fourth parallel was that boys were overrepresented in the classes with higher levels of ASD and/or ADHD, which corresponds with the upper extreme ASD and ADHD traits being more easily recognized in boys than in girls (Kramer, Krueger, & Hicks, 2008). This may indicate that sex differences at the upper extreme end of phenotypic traits are embedded in more typically developing children, as has been discussed previously (Neuman et al., 2005). This finding may suggest that sex differences in clinical referral and diagnoses of ASD and ADHD are not based on a clinical bias, but rather reflects a true predisposition in males. Hence, as recently discussed by Fair and colleagues (2012), our findings suggest that heterogeneity in clinical developmental disorders are rooted in comparable heterogeneity present in the general population. Given that previously more than 80% of our population sample was lumped together into “normal” classes (van der Meer et al., 2012), the current findings can be seen as an extension of those findings across the unaffected end of the ASD and ADHD traits.
Of particular relevance is the extension into the general population of the dissociation in visual-spatial functioning (block pattern performance and working memory) between classes displaying either more ASD traits than ADHD traits or vice versa, which we had found previously for clinical cases. This dissociation might pinpoint toward differential organization and/or functioning of neural substrates underlying visual-spatial information processes that are oppositely involved in ASD versus ADHD pathology. While normative visual-spatial attention is biased toward global processing (i.e., global interference), children with ASD are said to have a visual perceptual processing style that facilitates local rather than global processing (Booth, Charlton, Hughes, & Happé, 2003; Happé & Frith, 2006). Such local processing is favorable in the completion of cognitive tasks like the block pattern and embedded figures test (EFT), tasks on which individuals with ASD or high levels of mild ASD-like traits often show a superior performance since no global perceptual bias needs to be surpassed (Grinter et al., 2009; Shah & Frith, 1983). A reduced global-to-local perceptual process in ASD has also been approved in recent fMRI studies (Just, Keller, Malave, Kana, & Varma, 2012; Liu, Cherkassky, Minshew, & Just, 2011; McGrath et al., 2012), which may indicate less top-down control and increased local connectivity in ASD. Such increased local connections were not found in studies in ADHD; neural activity patterns rather indicated a frontal, striatal, and parietal hypofunction in ADHD (Bush, Valera, & Seidman, 2005; Silk et al., 2005; Vance et al., 2007). The widespread reduced top-down control described in those studies is not specific for visual processes in ADHD, and may pinpoint toward an overall reduced attentional network. We aim to follow-up on these findings by comparing children with elevated scores on either ASD or ADHD, children with elevated scores on both, and control children regarding brain activation patterns during visual-spatial task performance.
This study was not without limitations. First, information on all phenotypic domains relied on parent reports. Compared with clinical interviews, surveys tend to overestimate the degree of co-occurrence of ASD and ADHD, since the degree of response variability that can be measured is limited (Tourangeau, Rips, & Rasinki, 2000). This limitation may have affected the distribution of children across the classes in favor of the concordant classes and may have hampered the disclosure of discordant classes. Therefore, the optimal approach for studying these traits is the use of structured psychiatric interviews. Second, questionnaires measuring the symptom levels of the other internalizing and externalizing traits are not designed to examine the lower extreme end of the phenotypic spectrum. This however, did not impede differentiation across the classes, as can be seen in differences among the classes in other internalizing and externalizing problems (see also Table 3). Third, apart from the visual-spatial functioning, the classes did not differ on the other cognitive domains (see also Table 1). This may be due to the weakened associations between the cognitive measures on the one hand and the reduced range of scores on both ASD and ADHD traits on the other. Fourth, as aforementioned, sex differences were profound across all classes, with boys overrepresented in the classes with higher levels of the ASD (and ADHD) traits and girls overrepresented in the class with the lowest levels of both traits. However, we do not believe this has affected the results, since the effect of sex was analyzed and, when necessary, accounted for in the study.
In sum, the present study showed that, in the general population, 77.5% of the children presents with concordant ASD–ADHD trait profiles, while 22.5% of the children displays atypical discordant trait profiles characterized by differential visual-spatial functioning. This dissociation was previously also reported in classes with clinical symptoms of ASD and ADHD, suggesting that heterogeneity in ASD and ADHD is rooted in heterogeneity present in the unaffected end of the distribution.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by the Netherlands Organisation for Scientific Research (NWO) by grants assigned to Rommelse (91610024) and Buitelaar (056-13-015)