Sensitivity and Specificity of Original and Revised ADOS Algorithms in a Clinical Setting

Background: The Autism Diagnostic Observation Schedule (ADOS) was developed as an observational component of a diagnostic assessment for autism spectrum disorders (ASD) that includes a thorough history and other developmental tests.  Because administration is relatively brief relative to other ASD assessments, the ADOS has come into routine use for diagnostic classification in many clinical and educational settings. In 2007, the algorithms for Modules 1 - 3 were revised for improved comparability and measurement of severity across modules.
Objectives: The aim of this study was to examine the sensitivity and specificity of both original and revised ADOS algorithms when used as part of the standard clinical evaluation of children referred to a specialty center to rule out ASD.
Methods: Records of new visits to the Division of Developmental and Behavioral Pediatrics (DDBP) at Cincinnati Children's Hospital Medical Center in 2008 were reviewed.  Records of children referred for evaluation with ADOS Modules 1 - 3 were more closely examined for comparison of final clinical diagnosis to ADOS classification on the new and original algorithms. All ADOSs at Cincinnati Children's are administered by speech-language pathologists specializing in autism who have met and maintain research reliability standards on the instrument. Final clinical diagnosis is made by a developmental pediatrician following a multi-disciplinary team assessment. Final diagnosis and ADOS classification were collapsed into Spectrum (autism and ASD) and Not Spectrum for comparison in 2x2 tables.
Results: In 2008, approximately 2200 children were seen for the first time in DDBP for concerns about developmental delay. Of these, 657 underwent a multi-disciplinary evaluation to rule out ASD. A total of 603 children were evaluated with ADOS Modules 1 - 3. Complete ADOS and final clinical diagnosis information was available for 584: Module 1 No words  = 87; Module 1 some words = 90; Module 2 < 5 years = 107; Module 2 => 5 years = 91; Module 3 = 209. Final clinical diagnosis for all modules combined was Autism = 142 (24%); Non-autism ASD = 185 (32%) and Not spectrum = 257 (44%).  Sensitivities for both old and new algorithms ranged from 76 to 99 across the modules. Specificities were low for both old and new algorithms, but were consistently lower with the new algorithm. Specificity = 29% for Module 1 No words. The most common non-spectrum clinical diagnosis for children classified as ASD by the ADOS was global delay.  For Module 3, specificity = 34% and the most common non-spectrum diagnosis for children classified as ASD by the ADOS was anxiety/ADHD.
Conclusions: The ADOS provides valuable observational information to the diagnostic team in the clinicalevaluation of a child for ASD. However, in this clinical setting where children with numerous kinds of developmental disorders are evaluated, the specificity of the instrument is low and the information must be assessed as part of the overall evaluation by the multi-disciplinary team. In settings such as schools or private practitioners' offices where the ADOS has become the primary determinant of diagnostic classification, ASD may be over diagnosed.