Developmental Deviance of Item-Level Responses on Standardized Language Measures Correlates with Autism Spectrum Disorder Diagnosis

Background: Developmental deviance (DDEV) refers to the non-sequential attainment of milestones within a developmental domain. This observation is in contrast to developmental delay (DD), where milestones are reached in the typical sequence, but the timeline of attainments is delayed. There is evidence that DDEV is associated with certain neurobehavioral diagnoses, such as autism spectrum disorder (ASD). Clinically, the attainment of developmental milestones is assessed through standardized measures of developmental domains. Many psychometric tests are arranged hierarchically, and on the surface, two individuals with the same overall score on a clinical measure may appear to be impaired to a similar extent. However, at the itemized level, individuals with DDEV exhibit a more scattered pattern of incorrect answers. Differentiating between DDEV and DD may inform prognosis and predict long-term outcomes.

Objectives: We used a modified measure of scatter, called inefficiency, to differentiate between DD and DDEV using standardized measures of language ability. We tested the accuracy of inefficiency to predict ASD diagnosis, and by extension DDEV, among probands from the New Jersey Language and Autism Genetics Study (NJLAGS) cohort.

Methods: NJLAGS consists of 157 families (500 individuals) ascertained for at least one individual with ASD and another with language impairment (LI). All individuals were given the Clinical Evaluation of Language Functioning (CELF) and Comprehensive Assessment of Spoken Language (CASL). Four phenotypic groups were defined: unaffected, ASD, LI, and LI+ASD. A hierarchical clustering model was fitted to each subtest to investigate if clusters of item responses correlate with an individual’s diagnosis (LI or LI+ASD) regardless of scatter and ceiling effects. Inefficiency was defined for each subtest as the product of the total number of subtest items and the sum of the weights (percentage of unaffected family members who correctly answered the item) of the items missed. Group differences in inefficiency across subtests were assessed using ANOVA. We fit a generalized linear model to determine diagnostic outcomes of inefficiency with IQ and age included as covariates.

Results: Hierarchical clustering analysis indicated that the LI and LI+ASD groups segregate according to their item responses for subtests of the CASL and CELF. Overall raw scores did not differ between the LI and LI+ASD groups; however, the LI+ASD group had consistently higher inefficiency scores than the LI group for all language measures. The following subtests reached significance (p<0.001): CASL Nonliteral Language (NL), Pragmatic Judgement (PJ), Meaning From Context (MFC), and CELF Recalling Sentences (RS), Word Definitions (WD), and Word Classes (WC) subtests.  When controlling for age and IQ, inefficiency was able to accurately predict ASD diagnostic status among the LI and LI+ASD groups for the NL (p=0.002), PJ (p=0.009), MFC (p=0.005) and RS (p=0.005) subtests.

Conclusions: Individuals with LI+ASD exhibit more DDEV, as measured by inefficiency, across measures of expressive, pragmatic, and metalinguistic language compared to individuals with LI. By distinguishing between DDEV and DD, inefficiency was able to predict ASD diagnosis among LI/LI+ASD probands. Inefficiency can be applied to measures across multiple developmental domains in order to characterize developmental profiles of individuals with DDEV/DD.