Using latent profile analysis to evaluate the utility of a DSM-5 symptom checklist in screening children for autism spectrum disorder

Abstract

Objective: Currently available screening measures for Autism Spectrum Disorder (ASD) typically do not reflect DSM-5 diagnostic criteria and generally have weak positive predictive values. These factors result in missed opportunities for early intervention, delays in diagnosis, and contribute to inefficient usage of healthcare resources by inadequately discerning those in need of comprehensive assessment. This study examined a DSM-5 ASD symptom checklist to determine whether parent-report response patterns could accurately identify which children received an ASD diagnosis. Method: Data were examined from 376 ASD evaluations in a three-year period. Latent profile analysis was used to determine if subgroups could be identified according to parent response patterns, and network analysis was implemented to examine the relationship among DSM-5 ASD criteria within each latent profile. Results: A four-profile model was best supported based on fit indices and high probability classifications. The model was largely a product of how parents responded regarding their child’s sensory behavior and minimally reflected other symptomatology. Subsequent network analyses by profile indicated weak coherence among DSM-5 symptoms within all profiles. Overall, direct assessment of DSM-5 criteria based on parent report did not add diagnostic value beyond that reflected in base rates. Conclusions: Although continued refinement of ASD screening tools is needed to improve accuracy of referrals for evaluations and reduce wait time for diagnosis, this study continues to support the need for behavioral observation and formal assessment by trained clinicians. Continued development of sensitive and specific screening tools, likely with embedded behavioral and/or objective observation, is needed.