Trying to make sense of a heterogeneous disorder: A factor mixture modelling approach to autism spectrum disorder

Background: Autism spectrum disorder (ASD) is behaviourally defined by the presence of social and communication impairments and restricted and repetitive behaviours. The clinical phenotype is characterised by considerable heterogeneity, with individuals presenting with severe impairments through to more subtle deficits. This heterogeneity is often considered as a hindrance in the study of the aetiology and genetics of ASD.

Objectives: (1) To identify homogeneous subgroups within ASD using a Factor Mixture Modelling (FMM) approach. (2) To explore the similarities/differences of individuals assigned to each subgroup in terms of age, gender, diagnosis, and their symptom and cognitive profiles.

Methods: Participants were drawn from a large population-based sample of adolescent twins. The sample consisted of 251 individuals (M=13.5, SD=0.68; 174 males). 137 participants had a diagnosis of ASD (M=13.51 years, SD=0.73; 118 males), 40 had a diagnosis of broad spectrum autism (M=13.40 years, SD=0.56; 30 males), and 72 were unaffected co-twins (M=13.49 years, SD=0.65; 26 males). All twin pairs were behaviourally assessed for ASD symptomatology using parent report (Autism Diagnostic Interview-Revised, ADI-R) and direct observation (Autism Diagnostic Observation Schedule-Generic, ADOS-G). All twin pairs were also administered an extensive cognitive battery to measure IQ, language ability, theory of mind ability, executive functioning, and central coherence.

Analyses: FMM combines latent class analysis and confirmatory factor analysis to stratify individuals into relatively more homogeneous subgroups. Factor mixture models were tested using the raw subscale scores of the 37 items from the ADI-R. To guide the choice of the number of classes and factors for the FMMs, six latent class analyses (one-to-six classes) and three confirmatory factor analyses (one-to-three factors) were carried out. The fit of these models was assessed using goodness-of-fit criteria.

Results: Overall, a ‘two-factor, five classes’ FMM was chosen as the best fit of the data. According to this final FMM, individuals could be classified into five relatively homogeneous classes (C1: 23%, C2: 18%, C3: 17%, C4: 29%, C5: 13%, of the sample). A two factor solution fitted the data best; one factor corresponding to social/communication deficits, and a second factor corresponding to restricted and repetitive behaviour impairments.

Age did not differ across the five classes. There were a higher proportion of females in C1 and C2, with a higher proportion of males in C3, C4, and C5. The proportion of ASD diagnosis differed across the five classes, with C5 comprising only of ASD diagnoses. IQ also differed across classes.

Individuals assigned to C1 had the lowest social and communication impairments, with few restricted and repetitive behaviours. The severity of ASD symptoms significantly increased through C3 to C5, with individuals assigned to C5 showing the most severe impairments. However, the cognitive profile across the five classes was similar.

Conclusions: Five subgroups were identified using an FMM approach, which were largely based on symptom severity. Across the subgroups, individuals received different diagnoses, had a differing IQ profile, and a differing symptom profile. Age and cognitive profile was stable across subgroups. The findings also support the two symptom dimensions of ASD, as proposed in the DSM-5.