Abstract
Background: The Autism Diagnostic Interview, Revised (ADI-R) is a caregiver interview that is widely used as part of the diagnostic assessment for Autism Spectrum Disorder (ASD). Few large-scale studies have reported the sensitivity and specificity of the ADI-R algorithms, which are based on DSM-IV Autistic Disorder criteria. Kim and Lord (Journal of Autism and Developmental Disorders, 2012, 42, 82) developed revised DSM-5-based toddler algorithms, which are only applicable to children under 4 years. The current study developed DSM-5-based algorithms for children ages 4-17 years and examined their performance compared to clinical diagnosis and to the original DSM-IV-based algorithms.
Methods: Participants included 2,905 cases (2,144 ASD, 761 non-ASD) from clinical-research databanks. Children were clinically referred for ASD-related concerns or recruited for ASD-focused research projects, and their caregivers completed the ADI-R as part of a comprehensive diagnostic assessment. Items relevant to DSM-5 ASD criteria were selected for the new algorithms primarily based on their ability to discriminate ASD from non-ASD cases. Algorithms were created for individuals with and without reported use of phrase speech. Confirmatory factor analysis tested the fit of a DSM-5-based two-factor structure. ROC curve analyses examined the diagnostic accuracy of the revised algorithms compared to clinical diagnosis.
Results: The two-factor structure of the revised ADI-R algorithms showed adequate fit. Sensitivity of the original ADI-R algorithm ranged from 74% to 96%, and specificity ranged from 38% to 83%. The revised DSM-5-based algorithms performed similarly or better, with sensitivity ranging from 77% to 99% and specificity ranging from 71% to 92%.
Conclusions: In this large sample aggregated from US clinical-research sites, the original ADI-R algorithm showed adequate diagnostic validity, with poorer specificity among individuals without phrase speech. The revised DSM-5-based algorithms introduced here performed comparably to the original algorithms, with improved specificity in individuals without phrase speech. These revised algorithms offer an alternative method for summarizing ASD symptoms in a DSM-5-compatible manner.