Predicting Language in Children with ASD Using Spontaneous Language Samples and Standardized Measures

• In Event: Poster Session 14
  In Poster Session: PS 14 Topic - Developmental Disabilities

Abstract

Background: Previous research has indicated that standardized measures of language are often not sensitive to the heterogeneity of language abilities in children with Autism Spectrum Disorder (ASD) (Kasari et al., 2013). One suggestion has been the use of spontaneous language samples to capture a wider range of grammatical abilities in children with ASD that are not always apparent on standardized instruments (Wittke et al., 2017). The current project examines the degree to which standardized measures and natural language samples predict change in language skills in a longitudinal sample of children with ASD.

Methods: A subset of participants from the Autism Phenome Project (APP; N=54; 41 males) were selected for the current study. The APP is a longitudinal study intending to define clinically meaningful ASD subtypes based on behavioral and biological data. Participants were selected based on availability of video recordings of the Communication and Symbolic Behavior Scales (CSBS; Wetherby & Prizant, 2002) at initial visit (T1) and the Autism Diagnostic Observation Schedule (ADOS; Lord et al., 1999) three years later (T3). Coding of spontaneous language samples from CSBS and ADOS testing was used to calculate mean length of utterance (MLU) and total word types (TWT) at both T1 and T3. Mean age of participants at T1 was 33.9 months (SD=5.5) and 67.2 months (SD=10.2) at T3. In addition to MLU and TWT, formal testing at T1 from the Mullen Scales of Early Learning Expressive Language Raw Scores (MSEL-EL) as well as ADOS Social Affect (SA) and Restricted and Repetitive Behaviors (RRB) Total Scores were included as variables of interest.

Results: Two stepwise regression models evaluated predictors of language skills at T3; one for MLU and the other for TWT. T1 MSEL-EL (R2adjusted = .438, F(1,52) = 42.327, p<.001) and T1 MLU (ΔR2= .050, F(2,51) = 25.343, p=.029,) separately accounted for significant amounts of variance in T3 MLU. In contrast, T1 MSEL-EL(R2adjusted = .533, F(1,52) = 61.567, p<.001), T1 ADOS-SA (ΔR2 = .034F(4,49) = 19.298, p=.043), and T1 ADOS-RRB (ΔR2= .040, F(5,48) = 17.926, p=.024), each separately accounted for significant proportions of variance in T3 TWT.

Discussion: Both standardized language and autism testing, and spontaneous language samples were unique predictors of later language skills in this heterogeneous sample of children with ASD. MLU at T1 accounted for a unique proportion of the variance in later MLU at T3, whereas ADOS SA and RRB accounted for variance in later TWT at T3. Thus, early grammatical language use significantly predicts later grammatical language use (T1 and T3 MLU, respectively) over and above standardized MSEL-EL scores. Moreover, higher levels of autism symptomatology negatively impacted later vocabulary skills (indicated by lower TWT scores). One potential explanation is that social interaction skills facilitate the acquisition of new words, although not (apparently) grammatical skills. These results suggest that valuable predictive information can be obtained from both standardized measures and natural language samples in children with ASD; especially, spontaneous language scores should be considered when predicting longitudinal change in how children with ASD use language.

Authors

• Rebecca Podell Thomas, University of Connecticut
  Presenting Author

• Kacie Wittke, University of Connecticut
  Non-Presenting Author

• Jessica Elaine Blume, Texas Tech University
  Non-Presenting Author

• Ann M Mastergeorge, Texas Tech University
  Non-Presenting Author

• Letitia R Naigles, University of Connecticut
  Non-Presenting Author