Utility of the Child Behavior Checklist in Differentiating Children with Autism Spectrum Disorders from Other Clinical Disorders

Background:  

Identification of ASD is often challenging due to the heterogeneous presentation and symptom overlap with other clinical disorders (OD). “Best practice” diagnostic assessment of ASD is time- and resource-intensive, due to the need for in-depth assessment of functioning and behavior with autism-specific instruments by clinicians with specialized training (Huerta & Lord,2012). Thus, there is a pressing need for screening tools to identify children most in need of further ASD-specific assessment.

Recent reports have suggested that the CBCL (Achenbach & Rescorla,2001) may be useful in screening for ASD.

Three CBCL profiles in particular have been found to differentiate ASD from OD, including the Withdrawn (WD) subscale in both age versions, the Pervasive Developmental Problem(PDP) in the preschool version, and the “CBCL-ASD-Profile”(CAP) consisting of subscales WD, Social problems and Thought problems in the school version (Biederman et al.,2010;Narzisi et al.,2013,e.g.). However, studies examining their predictive validity have been limited by small samples of children with ASD, and comparison groups that have excluded children with cognitive impairments. Additionally, diagnostic accuracy may vary by gender and IQ level (Sikora et al.,2008). 

Due to widespread use of the CBCL as a routine screening instrument in mental health settings worldwide, reports of its usefulness in screening for ASD may lead clinicians to choose this over autism-specific screening tools. Thus, further study of its ability to correctly identify children with ASD is clearly warranted.

Objectives:  

Examine the utility of proposed CBCL profiles for ASD in differentiating children with ASD from OD, and potential differences by gender and nonverbal IQ (NVIQ).

Methods:  

Participants were 226 children with ASD and 174 children with OD (mainly ADHD/ODD, Anxiety/Mood disorder, and Intellectual Disability;ID), recruited as part of a research study. Diagnoses were made based on information from comprehensive assessments for ASD including questionnaires, ADI-R, Vineland-II, ADOS and cognitive testing.

Diagnostic accuracy for ASDdiagnosis was analyzed using ROC-analyses and area under the curve (AUC) scores. Analyses were done for the whole sample, and by gender and NVIQ (above/below 70). T-scores were used for WD and PDP, and the sum of raw scores of WD, Thought Problems and Social Problems for CAP. 

Results:  

Overall, the diagnostic accuracy for ASD was poor, with AUC at .67 for WD-preschool-aged, .66 for WD-school-aged, .67 for PDP, and .68 for CAP. At cutoffs low enough to obtain sensitivities at 80% (80-85%), specificities were low (38-48%).

Girls had better AUC-scores for PDP (.83 vs .61), WD-preschool-aged (.80 vs .63), and CAP (.72 vs .67), but not for WD-school-aged (.65 vs .67). For PDP, a cutoff of 61 correctly classified 95% of girls and 77% of boys with ASD, and 61% of girls and 44% of boys with OD.

ROC-analyses by NVIQ only showed better than poor AUC for CAP, which was moderate for higher IQs, but very poor for low IQs. 

Conclusions:  

The CBCL has limited utility in differentiating between children with ASD and OD such as ADHD/ODD, anxiety/mood disorders and ID. Our results do not support its use in place of well-established screening tools for ASD.