International Meeting for Autism Research: The Structure of Autism Symptoms as Measured by the ADOS

The Structure of Autism Symptoms as Measured by the ADOS

Friday, May 13, 2011
Elizabeth Ballroom E-F and Lirenta Foyer Level 2 (Manchester Grand Hyatt)
2:00 PM
M. Norris1 and L. Lecavalier2, (1)University of Rochester, Rochester, NY, (2)Ohio State University, Columbus, OH, United States

Several recent studies have examined the structure of autism symptoms (e.g., Frazier et al., 2008; Georgiades et al., 2007). This exercise can inform the instruments used to measure autism spectrum disorders (ASDs) as well as diagnostic classification systems. This is especially relevant with the impending release of DSM-V. Quantitative phenotypes also can help clarify the relationship between genes and behavior.

Thus far, research examining the structure of autistic symptoms has been inconclusive. There are important variations across studies in terms of sample characteristics and methodology. Most studies have relied on the ADI-R.  Overall, studies support either two- or three-factor solutions, though the composition of these factors varies across studies.


The primary objective was to compare different models of autism symptoms within a large sample of individuals with ASDs using the ADOS.  Three models were compared: a one-factor model; a model based on the DSM-IV diagnostic criteria; and a two-factor model proposed by the DSM-V committee. In the latter model, one factor consisted of social and communication behaviors and the other factor consisting of restricted, repetitive behavior and language items. A secondary objective was to examine the impact of age and level of functioning on model fit.


Participants included individuals aged 3-18 years (N = 1,409) recruited from the Autism Genetic Resource Exchange (AGRE) database. Participants were excluded if they did not have an ASD diagnosis, diagnosis was unclear, or if spectrum cut-offs on the ADOS were not exceeded. Modules 1 and 3 of the ADOS were examined.

Confirmatory factor analysis was performed using LISREL. Polychoric correlation matrices were used due to the ordinal nature of the data. Various indices of fit (RMSEA, SRMR, NNFI, CFI, and GFI) were compared across models, as well as across subsamples within models. RMSEA was used as the primary index. 


Results indicated that models were very similar to each other. Within Module 1, all three models fit well (RMSEAs ranged from .056 to .062). Within Module 1, model fits improved when analyses where conducted on subsamples based on age (those ≤6 years) or level of functioning (VABS composite and DLS  <55); RMSEAs for subgroup analyses ranged from .04 to .059. Models did not fit as well with Module 3 data, but again were similar to each other (RMSEAs ranged from .074 to .083). Within Module 3, indices of fit improved when analyses were conducted on a subgroup of older children (those ≥10 years); RMSEAs ranged from .068-.079. Fits worsened when analyses were conducted on lower and higher functioning subsamples (based on VABS composite and DLS cut-off of 70); RMSEAs ranged from .078 to .148.  


Generally, there was little differentiation between models. Indices of fit were impacted by sample characteristics, though not always as anticipated. There were notable differences between Modules 1 and 3, which may be further evidence that symptom structure changes with development. Multivariate statistics are but one method to study autism symptoms; other methods will be necessary.  

| More