Examining the Autism Phenotype

Autism spectrum disorders (ASDs) are widely studied yet poorly understood. They are characterized by impairments in social and communication skills and the presence of repetitive and stereotyped behaviors and/or circumscribed interests (RRBs). Much recent attention has been directed towards elucidating the structure of autistic symptoms. A better understanding of the phenotype can lead to improved diagnoses and clarification of etiology and pathogenesis. Factor analytic studies are one way researchers have pursued this end. Most often, two- or three-factor solutions have been reported. The objective of the current study was to test several competing models of the autism phenotype using theAutism Diagnostic Observation Schedule (ADOS). Participants included individuals with ASDs aged 3-18 years (N = 1,409) from the Autism Genetic Resource Exchange database. ADOS data from 720 participants for Module 1 and 689 participants for Module 3 were used in analyses. Samples were divided into more homogenous subgroups to examine the impact of age and level of functioning on model fit. Confirmatory factor analyses (CFAs) were performed on total samples and subsamples. Four primary models were tested: (a) a one-factor model; (b) a two-factor model (one factor consisting of social/ communication items and the other consisting of RRBs); (c) a three-factor model based on DSM-IV-TR criteria (social, communication, and RRB factors); and (d) a two-factor model based on proposed DSM-V criteria (one factor consisting of social/ communication symptoms and one factor consisting of restricted and repetitive behaviors and language). Additionally, ADI-R RRB items were added to analyses because these symptoms may not be well captured with the ADOS. Bi-factor models were also examined for the DSM-IV analyses in order to examine the possibility that ASD symptoms were best explained by one general ASD factor and three specific factors. Results of the CFAs with Module 1 indicated all models fit reasonably well, with RMSEAs ranging from .056 (DSM-IV model) to .062 (one-factor and two-factor models). RMSEA confidence intervals overlapped, suggesting no model fit significantly better than other models. Generally, fit improved in the analyses with more homogenous subgroups. Addition of ADI-R RRB items resulted in un-interpretable results. Results of the CFAs with Module 3 indicated unacceptable fit for most models, with RMSEAs ranging from .074 (DSM-V) to .083 (one-factor model). RMSEA confidence intervals again overlapped, indicating all models fit similarly. Unlike Module 1 analyses, indices of fit improved when ADI-R RRB items were included in analyses, but there was again little differentiation between models. Fit improved in the analyses with more homogenous subgroups by age, but not level of functioning. Finally, the bi-factor DSM-IV model did not aide interpretation in either module. The lack of differentiation between models in both modules suggests that the structure of ASD symptoms is complex and several research methods will be necessary to understand the symptom structure. It may also help explain why different solutions are found across studies; that is, models are similar to each other and different fit indices are found with different subgroups.