Deconstructing the PDD Clinical Phenotype: Internal Validity of the DSM-IV

Background: Current diagnostic and classification systems conceptualize pervasive developmental disorders (PDDs) in terms of impairments in social interaction, communication, and repetitive-restricted behaviors. Empirical studies of the structure of PDD symptoms have challenged this three domain model of impairment. For example, some authors have proposed a single dimension to characterize PDD symptoms, others have proposed two-dimensional models consisting of social-communicative and repetitive-restricted behaviors, and yet others have proposed models with more dimensions.

Objectives: To assess the internal validity of the DSM as a conceptual model for describing PDD, while paying particular attention to certain subject characteristics. In order to overcome methodological shortcomings of previous studies, we examined DSM-IV symptoms from all three core PDD dimensions, included a large sample of PDD children with the full range of symptom severity, and assessed competing models with confirmatory factor analysis (CFA). We examined the impact homogeneous subgroups (based on age, IQ, and PDD subtype) had on model fit.

Methods: Parents and teachers completed a 12-item DSM-IV-referenced rating scale for 3-to-12 year old clinic referrals with a PDD. The sample consisted of 730 children aged 3 through 12 years (mean=7.1; SD=2.5) and was predominantly male and Caucasian. Using DSM-IV criteria, clinical diagnoses were made by an expert clinician. All children met DSM-IV criteria for autistic disorder, Asperger’s disorder, or PDDNOS. Ratings were submitted to CFA and different models were assessed for fit. Analyses were first conducted on the entire sample and then based on age (preschool, n=229 vs school age, n=501), PDD subtype (autism n=254, Asperger’s, n=154, and PDDNOS, n=322), and in 6-12 years olds with IQ>70 (n=263). Three models were tested in every subsample: a one-factor model, a two-factor model consisting of the eight social-communication items and four restricted/repetitive behavior items, and a three-factor model representing the DSM-IV triad of impairments.

Results: Measures of fit indicated that the three-factor solution based on the DSM was superior to other models. The one- and two-factor models yielded poor fit for both informants. Fit indices varied according to the rater, child’s age, PDD subtype, and IQ. Overall, factor loadings were high for both parents and teachers. Subgroup analyses impacted indices of fit, especially for parent data. For instance, models for school age children had better fit than observed with preschoolers (RMSEA=.075 vs .098) and improved if only children with IQ>70 were included in the analyses (RMSEA=.057). In terms of PDD subtypes, fits were best for Asperger’s disorder (RMSEA=.039) and poorest for autism (RMSEA=.095). There was less differentiation across subject characteristics with the teacher data.

Conclusions: Results clearly indicated that the three-factor solution provided better fit to the data than the one- and two-factor solutions previously reported in the literature. Regardless of subject characteristics, the three-factor solution always resulted in better fit indices than the other models tested. The current data also showed that subject characteristics impacted fit. More research needs to be done before discarding current classification systems. Subject characteristics, modality of assessment, and procedural variations in statistical analyses impact conclusions on the structure of PDD symptoms.