Item Response Theory Analysis Suggests Word-Level Lexical Differences in Children with Autism at 12 Months of Age: A Multisite Infant-Sibling Study

Background:  Parent-report communication inventories have found that language development in toddlers with autism is similar to that of language-matched controls in terms of semantic categories, common nouns, predicates, and closed-class terms (Charman et al. 2003, Weismer et al. 2011, Rescorla and Safyer 2013). Two studies have analyzed word-level lexical development in toddlers with autism. While Rescorla and Safyer (2013) found similar word-level production, Bruckner et al. (2007) found 25 words were either more or less likely to be understood by toddlers with autism.

Objectives:  To determine if word-level lexical differences in production and understanding could be detected at 12 months of age in infants who later developed autism, compared to infants who did not.

Methods:   The 346 participants included in this cross-sectional analysis were part of four larger longitudinal studies of infants with an older sibling either with ASD (high risk – HR) or without (low risk – LR).  43 HR siblings were later diagnosed with ASD at 36 months based on the Autism Diagnostic Observation Schedule and independent review/clinical confirmation.  170 HR and 133 LR siblings did not receive a later ASD diagnosis. Item Response Theory was applied to determine the relative probability of understanding and producing each of the 396 words on the McArthur-Bates Communicative Development Inventories (CDI).  Word-level Differential Item Functioning (DIF) was calculated to determine which words were more or less likely to be understood or produced by the autism outcome infants when equated for language ability (theta).

Results: At 12 months of age, autism outcome infants were more likely to use words in a statistically unexpected or improbable manner (producing certain less probable/more difficult words earlier, relative to their overall level of lexical development).  The mean number of words produced by the non-ASD outcome HR and LR infants was five.  The HR infants with later ASD produced a mean of 2.7 words and a total of 23 different words.  Six of these words, or 26 percent of the total, were produced in a statistically unique way by the infants with later ASD (DIF>1.0 and p<0.05). Relative to their overall language ability on the CDI, they were more likely than expected to produce "baabaa," "block," and "bye" when compared to LR infants and "baabaa,” “uncle," "hello," and "bite" when compared to HR infants without later ASD.  No words for the infants without later ASD were statistically unexpected. A similar pattern was found with words understood.

Conclusions:  Qualitative language differences were measured at 12 months of age on the CDI in infants who later developed autism. The interpretation of word-level DIF is limited because the instrument did not reveal a word's accuracy, context, frequency, or range of extension. The differences may reflect 'early' linguistic manifestations of reduced social communication, restricted/repetitive interests or impaired language.  Next Step: Analyze 18-month CDI when there is more variability in the measure.