Validation of Proposed DSM-5 Criteria for Autism Spectrum Disorder

Background: Proposed DSM-5 criteria posit a single autism spectrum disorder (ASD) category with two symptom criteria sets – social communication and interaction and restricted, repetitive behavior. Recent work from our group has found a latent categorical distinction between ASD and typical autism symptom levels and numerous studies support a broad range of autism symptom severity spanning both social communication and repetitive behavior domains.

Objectives: The primary purpose of the present study was to examine whether parent-reported symptom data from a large internet registry support the validity of the structure of proposed DSM-5 diagnostic criteria. The study also evaluated whether proposed DSM-5 criteria showed enhanced sensitivity and specificity to autism relative to DSM-IV-TR criteria.

Methods: Data were obtained from the Interactive Autism Network (IAN; N=14,744). IAN preferentially recruits families with at least one affected child who has been diagnosed with an ASD. Caregivers reported autism symptoms using the Social Responsiveness Scale (SRS) and the Social Communication Questionnaire (SCQ). Exploratory factor (FA), latent class (LCA), and factor mixture (FM) models were computed for each measure and across the total sample and a parent-designated unaffected sibling sub-sample. Empirical classifications from FM models were compared to parent-designated affected status. Additionally, SCQ and SRS symptoms were mapped onto specific DSM-5 and DSM-IV-TR criteria. DSM-5 and DSM-IV-TR criteria were then used to predict parent-designations and empirical classifications of ASD using diagnostic efficiency statistics.

Results: Two-factor/two-class FM models showed superior fit to FA and LCA models and replicated across measures and in the total sample and unaffected sub-sample. The two factors were represented by social communication and repetitive behavior indicators. Factor means and variances supported the presence of autism and non-autism classes with broad, overlapping symptom distributions. Classes closely mirrored parent-designated affected and unaffected status (κ=.78; classification % overlap=89%). Social communication and repetitive behavior factors showed strong correlations in the autism-affected (r=.76) and unaffected groups (r=.84). Proposed DSM-5 criteria showed superior specificity relative to DSM-4 criteria (.97 vs. .87), however sensitivity was weaker (.81 for DSM-5 vs. .95 for DSM-IV-TR). Relaxing DSM-5 social communication criteria to 2 of 3 criteria met instead of the proposed 3 of 3 criteria increased sensitivity (relaxed .88 vs. proposed .81), without substantively reducing specificity (relaxed .96 vs. proposed .97).

Conclusions: The validity of DSM-5 criteria was supported, with the exception that requiring only 2 of 3 social communication criteria may enhance sensitivity with minimal decrease in specificity. The strong correlation between social communication and repetitive behavior domains in the autism category supports the requirement for symptom criteria to be met in both domains. A categorical autism distinction reinforces the need for an evidence-based medicine approach to ASD diagnosis where post-test probabilities are iteratively generated using the population or clinical base rate and non-redundant predictors of ASD diagnosis. The lack of additional sub-categories within ASD substantiates the validity of a broad categorical diagnosis containing two sub-dimensions. Genomic and neurobiological research may benefit by using mixed models that accommodate both the categorical and dimensional nature of ASD symptoms.