Sensitivity and Specificity of the ADOS 2 Algorithms in a Large German Sample

Background:

The Autism Diagnostic Observation Scale (ADOS) in combination with the Autism Diagnostic Interview-revised is considered to be the gold standard for

diagnosing Autism Spectrum Disorder (ASD) and Autism. Recently, a new algorithm (ADOS-2) has been proposed with the goal to achieve better sensitivity and specificity

as well as comparability between modules. The description of a new algorithm for module 4 followed in March 2014.

Objectives: Here, we compared sensitivity and specificity of the original ADOS and the new ADOS-2 algorithms in a sample of 448 child psychiatric patients to replicate

ADOS-2 validity in a German sample.

Methods:

447 dataset (modules 3 and 4) were re-evaluated using the revised ADOS-2 algorithms. Diagnosis (BEC, best estimated clinical diagnosis) was established

after the administration of ADOS, ADI-R, SCQ, SRS, CBCL, and IQ-tests by an experienced and independent clinician (psychologist or psychiatrist) according to ICD-10

. N=233 children had a diagnosis of autism, N=124 children had a diagnosis of Asperger Syndrome or atypical autism (Spectrum Diagnosis), and in N=87, ASD was ruled

out, but another psychiatric diagnosis was confirmed. Statistical analysis: Diagnostic validity was examined for module 3 (N=315) and module 4 (N=132) by comparing the

best estimate clinical diagnosis to the results of the

ADOS and ADOS 2 algorithm by receiver operating characteristics curve (ROC). Test accuracy was measured by AUC, and sensitivity and specificity for autism versus

non-ASD and ASD versus non-ASD were calculated. We used the original cut-offs described for the American samples.

Results:  

Age at the time of diagnosis ranged from 5 to 16 years for module 3 (mean 10.2. SD 2.1) and from 12 to 40 years for module 4 (mean 18.0. 4.9). For module 3, 89% and

for module 4, 83% of the participants were male. IQ total scores ranged from 41 to 147 for module 3 (mean 98.1. SD 18.4). For module 4, IQ total scores ranged from 44

to 141 (mean 91.3, SD 18.1). For module 3 (non-ASD vs ASD) ADOS AUC was .87 (CI 0.82-0.92) and ADOS-2 .87 (CI 0.82- 0.92). For non-ASD vs Autism ADOS AUC was .91

(0.86-0.95) and ADOS-2 AUC was .90 (CI 0.86-0.95). For module 4 AUC for ADOS algorithm non-ASD vs ASD was .87 (CI 0.79-0.95) and .82 (CI 0.73-0.91) for ADOS 2

algorithm. For non-ASD vs Autism AUC was .92 (CI 0.87-0.98) and for ADOS-2 .87 (CI 0.8-0.95).

Conclusions:

For modules 3 and 4 we found comparable results for the original ADOS and the new ADOS-2 algorithms in this clinical German sample, differentiating

non-ASD patients from ASD and autism patients. The module 3 ADOS-2 algorithm showed for autism versus ASD higher sensitivity and a slight decrease in specificity

compared to ADOS. These results replicate earlier findings concerning the continued validity of ADOS as a diagnostic measure. For module 4 we found similar results as in

the original American sample with a good sensitivity and a slight loss of specificity for ADOS-2. Module 1 and 2 will be evaluated in the next step.