Research PaperA cluster analysis exploration of autism spectrum disorder subgroups in children without intellectual disability
Introduction
Autism spectrum disorder (ASD) is defined by impaired social communication skills and restricted, repetitive patterns of behaviour, interests, and activities (RRBI) (American Psychiatric Association, 2013). There is, however, significant variability within the ASD phenotype across behavior, development, and neurocognition. This results in diagnostic inexactitude and the inability to identify potential biomarkers of the disorder.
DSM-IV-TR defined three subgroups of ASD, namely, autistic disorder (AD), Asperger’s disorder (AS), and pervasive developmental disorder-not otherwise specified (PDD-NOS) (American Psychiatric Association, 2000). This led to diagnostic inconsistencies (Lord, Petkova et al., 2012), and the distinction between subtypes was unstable longitudinally (Woolfenden, Sarkozy, Ridley, & Williams, 2012). A further debate, not reflected in the DSM classification, has been a possible distinction between AD without intellectual disability (ID; often termed ‘high-functioning autism’ (HFA) in research literature), and AS based on differences in early language development (Bennett et al., 2008, Mayes et al., 2001). In the clinical setting, reliably diagnosing HFA and AS was hampered by difficulties establishing early language development retrospectively (American Psychiatric Association, 2010). Early language developmental milestones were not a reliable differentiator of profiles of neuropsychological functioning (Ghaziuddin and Mountain-Kimchi, 2004, Miller and Ozonoff, 2000), language ability (Lewis, Murdoch, & Woodyatt, 2007; Mayes & Calhoun, 2001), or developmental trajectories (Starr, Szatmari, Bryson, & Zwaigenbaum, 2003; Szatmari et al., 2000), questioning the clinical utility in differentiating HFA and AS based on the existing criteria.
In DSM-5 (American Psychiatric Association, 2013), ASD classification moved from subtypes to the notion of a spectrum disorder. The proposal is of two core dimensions that characterise ASD, which vary in composite symptoms and severity across individuals. In contrast to the DSM-IV-TR categorical system, all individuals are classified within a single diagnostic group. While this approach eliminates the previous confusion in diagnosing subtypes, there is concern that it minimises the phenotypic heterogeneity in ASD, particularly with individuals without ID (‘ASD without ID’). Distinguishable phenotypes within ASD without ID may exist, but it is likely that the previous diagnostic criteria did not adequately delineate the defined subtypes (Ghaziuddin, 2010). Moving forward using a more dimensional approach, individual differences can be determined by characterising symptom profiles. Data that examine such differences in an objective manner is a potential way forward to assess the utility of subtypes.
Cluster analysis provides an objective method of exploring the way that phenotypic characteristics group together. Increasingly, such exploratory techniques are being employed to reduce the phenotypic heterogeneity of ASD into more cohesive subgroups based on selected clinical features. Surprisingly, few cluster analytic studies have explored the variability in ASD symptomatology in children or adolescents with ASD without ID (Beckett, 2005; Bitsika, Sharpley, & Orapeleng, 2008; Prior et al., 1998, Verte et al., 2006).
In two cluster analytic studies of individuals with ASD without ID, three-cluster solutions were identified that loosely aligned with DSM-IV-TR classifications (Prior et al., 1998, Verte et al., 2006). On closer examination, however, the subgroups were primarily differentiated by the severity of ASD symptoms (Verte et al., 2006), or by variability in cognitive, communicative, and behavioural difficulties (Prior et al., 1998). Thus, the subgroups showed a similar profile differentiated by severity of impairment, not quality or type of ASD symptomology. These authors interpreted the findings as supporting the spectrum approach of ASD, as per DSM-5. When employing a data driven approach to explore potential subgroups, however, clusters will differ according to the variables analysed. Both Prior et al. (1998) and Verte et al. (2006) only sampled ASD symptomatology via parent report, which may have provided a biased perspective. Further, the ability to capture the heterogeneity of this population was limited by only analysing the presence or absence of symptoms (Prior et al., 1998), or by solely examining symptom domain scores (Verte et al., 2006). The sample of clinical variables may therefore have been limited and impacted the ability to reveal clinically meaningful subgroups.
Bitsika et al. (2008) did not limit their analysis to ASD symptoms; rather, they examined ASD severity, together with other functional indices, including cognition and adaptive functioning. In doing so, three clusters that differed significantly in reciprocal social interaction, communication, and adaptive functioning were described. The subgroups differed significantly in both the severity and profile of symptoms across core domains, supporting the potential to differentiate qualitatively distinct clusters. Thus, characterisation of core symptomatology together with associated clinical features may help to capture the phenotypic heterogeneity in ASD.
More recently, latent profile analysis has been used to examine the dimensional profile of ASD features. In a childhood ASD study including individuals both with and without ID, Greaves-Lord et al. (2013) identified six phenotypic classes when parent reported ASD symptomatology was examined. Classes 1–3 were reported to align with the DSM-5 conceptualisation of ASD and were characterised by different degrees of impairment within both social communication and RRBI domains. In contrast to classes 1–3, the authors indicated that classes 4–6 were primarily represented by children without ID and were less consistent with the DSM-5 concept of ASD. Class 4 was characterised by severe resistance to change, with comparatively low levels of stereotyped behaviours and social communication deficits; in contrast, Class 5 displayed moderate difficulty understanding social information and communication but low impairment across RRBI subdomains. Class 6 was subclinical on all ASD domains and was characterised by attention and disruptive problems. These findings suggest that the diagnostic complexity of ASD without ID may not be adequately captured by DSM-5.
Given the complexity and variability of the ASD phenotype, it is surprising that cluster analytic studies examining ASD without ID to date have commonly relied on either parent or child assessment measures to characterise functioning, without adopting both assessment techniques together in the single study. The Autism Diagnostic Interview - Revised (ADI-R; Rutter, Le Couteur, & Lord, 2003) caregiver interview, and Autism Diagnostic Observation Schedule - Second Edition (ADOS-2; Lord, Rutter et al., 2012) child assessment are the ‘gold standard’ assessment measures for ASD diagnosis. Exploration of potential childhood subgroups within ASD without ID using these instruments together may better evaluate phenotypic subgroups.
ASD without ID formed the focus of this study, where phenotypic variability and diagnostic uncertainty seems greatest. Within an Australian childhood sample, cluster analysis was used to explore the presence of phenotypic subgroups. We utilised the ‘gold standard’ ASD diagnostic measures, the ADI-R and ADOS-2, to provide a comprehensive evaluation of ASD characteristics. Cognitive, language, behavioural and emotional functioning was also examined. Understanding phenotypic profiles within this population is important for diagnostic classification systems, as well as clinically and educationally, to inform management strategies.
Section snippets
Participants
Sixty-one children (51 males; 10 females) aged 5–14 years (M = 8.81, SD = 2.36) participated with their primary caregiver. Child participants had a parent reported clinical diagnosis of DSM-IV-TR AD, AS, or PDD-NOS (American Psychiatric Association, 2000), or DSM-5 ASD (American Psychiatric Association, 2013); all diagnoses henceforth combined and referred to as ASD. Twenty-two participants had previously been diagnosed by a multidisciplinary group involving author RT, a paediatrician, and a speech
Cluster analysis
A two-cluster solution was optimal according to the Calinski Harabasz Criterion when current (ADOS-2) and lifetime (ADI-R) developmental and behavioural variables were subject to complete linkage cluster analysis. Consistent with this result, a two-cluster solution was also optimal when ADOS-2 and ADI-R data were examined independently. All three solutions were shown to be stable using bootstrapping techniques. The Rand Index examined concordance rates between the solutions. As expected, the
Discussion
This exploratory study described two clusters within ASD in children without ID based upon the pattern of severity across ASD dimensions. Results indicate that ASD without ID can be differentiated into Severe (Cluster 1) and Moderate Social Impairment (Cluster 2) subgroups when ASD features are more closely examined. The Severe Social Impairment subgroup displayed greater impairment in social interaction and communication skills, but lower lifetime severity of RRBI. In contrast, the Moderate
Conflict of interest
The authors declare there are no conflicts of interest.
Acknowledgments
We thank the families and teachers who generously volunteered their time for this study. Author C Pantelis was supported by a National Health and Medical Research Centre Senior Principal Research Fellowship (ID numbers: 628386, 1105825).
References (48)
- et al.
Linguistic abilities in children with autism spectrum disorder
Research in Autism Spectrum Disorders
(2007) - et al.
Relations among restricted and repetitive behaviors, anxiety and sensory features in children with autism spectrum disorders
Research in Autism Spectrum Disorders
(2014) - et al.
A systematic review of the diagnostic stability of autism spectrum disorder
Research in Autism Spectrum Disorders
(2012) Diagnostic and Statistical Manual of Mental Disorders
(2000)- American Psychiatric Association (2010). DSM-5 Development. Retrieved from...
Diagnostic and Statistical Manual of Mental Disorders: DSM-5
(2013)Cluster analysis of the putative diagnostic distinctions between Asperger syndrome and high-functioning Autism
Dissertation Abstracts International: Section B: The Sciences and Engineering
(2005)- et al.
Differentiating autism and Asperger syndrome on the basis of language delay or impairment
Journal of Autism and Developmental Disorders
(2008) Children’s Communication Checklist-2
(2003)- et al.
An exploratory analysis of the use of cognitive, adaptive and behavioural indices for cluster analysis of ASD subgroups
Journal of Intellectual Disability Research
(2008)
Exploring the fractionation of autism at the cognitive level
Autism: The International Journal of Research and Practice
A dendrite method for cluster analysis
Communications in Statistics
Restricted and repetitive behaviours, sensory processing and cognitive style in children with autism spectrum disorders
Journal of Autism and Developmental Disorders
Statistical power analysis for the behavioral sciences
Factor analysis of restricted and repetitive behaviors in autism using the Autism Diagnostic Interview-R
Child Psychiatry and Human Development
Functional magnetic resonance imaging of autism spectrum disorders
Dialogues in Clinical Neuroscience
Cluster analysis
Defining the intellectual profile of Asperger Syndrome: A comparison with high-functioning autism
Journal of Autism and Developmental Disorders
Brief report: Should the DSM V drop Asperger syndrome?
Journal of Autism and Developmental Disorders
Empirically based phenotypic profiles of children with pervasive developmental disorders: Interpretation in the light of the DSM-5
Journal of Autism and Developmental Disorders
The ‘fractionable autism triad’: A review of evidence from behavioural, genetic, cognitive and neural research
Neuropsychology Review
Time to give up on a single explanation for autism
Nature Neuroscience
Categorical and dimensional structure of autism spectrum disorders: The nosologic validity of Asperger syndrome
Journal of Autism and Developmental Disorders
Examining the relationship between executive functions and restricted, repetitive symptoms of autistic disorder
Journal of Autism and Developmental Disorders
Cited by (23)
Developmental and behavioral groupings can predict changes in adaptive behavior over time in young children with neurodevelopmental disorders
2023, Research in Developmental DisabilitiesCitation Excerpt :=0.00 Diagnostic labels for NDDs determined in early childhood are not necessarily good predictors of later outcomes in terms of day-to-day functioning, learning, or inclusion in various setting such as school or daycare (Klopper et al., 2017). Indeed, comorbid conditions or the emotional and behavioral challenges that tend to accompany NDD diagnoses can negatively impact these outcomes (Gillberg et al., 2014; Fernell et al., 2014; Miller et al., 2016).
Validation strategies for subtypes in psychiatry: A systematic review of research on autism spectrum disorder
2021, Clinical Psychology ReviewCitation Excerpt :Demographic variables may for example be age, sex, and level of education. Psychological variables may for example be questionnaires, cognitive tests (McCrimmon, Schwean, Saklofske, Montgomery, & Brady, 2012), or symptom checklists (Klopper, Testa, Pantelis, & Skafidas, 2017). Biological variables may for example be gene expression measurements (Kong et al., 2013), facial features (Obafemi-Ajayi et al., 2015), or EEG measures (Hasenstab, Sugar, Telesca, Jeste, & Şentürk, 2016).
Spectrum or subtypes? A latent profile analysis of restricted and repetitive behaviours in autism
2019, Research in Autism Spectrum DisordersCitation Excerpt :This is surprising, given the varied presentation of RRBIs among individuals on the spectrum, and indicates that cross-sectional phenotypic data may not provide enough information to define unique subtypes of autism. Previous research has identified groups of individuals with distinct profiles (e.g. Cholemkery et al., 2016; Greaves-Lord et al., 2013; Klopper et al., 2017). However, the ‘subtypes’ reported in these studies tend to only show disparities in symptomatic presentation (i.e. in social/communication and RRBI presentation) and not in other clinical measures (e.g. problematic behaviours, adaptive behaviours), which places limitations on their clinical utility.
Cluster analysis reveals subclinical subgroups with shared autistic and schizotypal traits
2018, Psychiatry ResearchCitation Excerpt :Although factor analysis quantifies the expression of individual factors, it does not group individuals according to similar characteristics. Considering the heterogeneity of ASC and SSC, and their symptom similarities, the identification of specific symptom subgroups within these two spectra may provide a helpful framework for investigating their neural and genetic correlates, and the development of more targeted interventions (Fonseca-Pedrero et al., 2017; Insel, 2010; Klopper et al., 2017). Cluster analysis groups individuals into homogenous clusters, according to their expression of certain traits, such as the above-mentioned factors.
Empirically-identified subgroups of children with autism spectrum disorder and their response to two types of cognitive behavioral therapy
2021, Development and PsychopathologyCognitive-perceptual traits associated with autism and schizotypy influence use of physics during predictive visual tracking
2023, European Journal of Neuroscience
- 1
Present Address: Developmental Disability and Rehabilitation Research, Murdoch Childrens Research Institute, The Royal Children’s Hospital, Flemington Rd, Parkville, Melbourne, Victoria 3052, Australia.