A Comparative Analysis of Three Autism Spectrum Disorder Screening Measures in a Clinical Population

Background: With the increasing prevalence of autism spectrum disorders (ASD), several instruments have been developed to assess and screen for characteristic behaviors and communication deficits in clinical populations. These measures have been used as level two screeners in clinical settings, but few studies compare the ability of the these instruments to reliably discriminate between ASD and non-spectrum (NS) diagnoses in a diverse clinical sample.

Objectives: To assess the ability of the Social Responsiveness Scale (SRS), Social Communication Questionnaire (SCQ) and Child Behavior Checklist (CBCL) to distinguish between ASD and NS disorders in a diverse patient sample recruited across clinical sites.

Methods: Forty-six subjects between 33 months and 17 years of age (M=83.24 months, SD=46.5) were recruited for a genetics study through three Children’s hospital departments (Outpatient Child Psychiatry=7, Developmental Services (speech, occupational therapy, developmental evaluation clinic)=10, and an autism specific clinic=18) and from community events and online resources (n=11). Diagnoses included autism=17, pervasive developmental disorder-not otherwise specified (PDD-NOS)=4, Asperger’s syndrome=2, attention deficit hyperactivity disorder=6, developmental delays=6, speech disorders=3, and other (anxiety, depression, genetic disorders)=8. Parents/legal guardians completed the CBCL, SRS, and SCQ. On the SCQ, a cut-off score of ≥12 was used for children <5 years and ≥15 for children >5 years.

Results: The mean SRS T-scores were significantly higher for the ASD (M=80, SD=12.65) than the NS group (M=68.64, SD=12.78), F(1,41)=8.58, p<.01.The mean SCQ score was also significantly higher for the ASD (M=17.62, SD=6.86) than the NS group (M=10.27, SD=6.36), F(1,41) =13.79, p<.001. However, categorical discrimination between diagnostic groups was much better for the SCQ (sensitivity=76%; specificity=77%) than the SRS.The SRS captured most children with ASD (sensitivity=90%) but misclassified most NS children when using a mild/moderate or severe categorization as the criteria for ASD (specificity=14%). When considering only the SRS severe classification as ASD, performance was similar to the SCQ (specificity=77%; sensitivity=71%).
Assessing discriminative validity of the SRS by using receiver operating curve analysis, the area under the curve was .75.Using a T-score of 73 rather than 60 resulted in a specificity of 73% and sensitivity of 76%. Age effects were found for the SCQ, and expanding the lower cut-off of >12 for children <8 years as suggested in an earlier paper (Corsello et al., 2007) improved sensitivity to 81%, but at the expense of specificity (68%). As expected, the CBCL autism scales (Withdrawn, Social Problems and PDD), were not as effective at discriminating between diagnostic groups (sensitivity=62%; specificity=41%).

Conclusions: Both the SRS and SCQ are widely used screening measures within clinical populations; however, age effects and the target populations must be considered when using them to screen for the presence of an ASD. When identifying children with ASDs in a clinical sample, using a lower cut-off for children <8 years on the SCQ and the severe classification on the SRS improved the combination of sensitivity and specificity for the measures.