Predicting and Modeling Distinct Developmental Trajectories of Adaptive Behavior during Pre-School to School-Age in Children with ASD

Background:  Adaptive behavior, frequently measured by the Vineland Adaptive Behavior Scales (VABS-II; Sparrow et al., 2005), is defined as the ability to function adequately and independently in the environment. Data on the developmental trajectory of adaptive behavior generally show that individuals with ASD do not gain skills at an age-appropriate rate (e.g., Anderson et al., 2009), and that increasing impairments are most strongly related to IQ (e.g., Szatmari et al., 2009) and more variably to ASD symptom severity (e.g., Kanne et al., 2010; Perry et al., 2009; Klin et al., 2007). However, samples used have been small, cross-sectional, or not well-characterized. Baghdadli et al. (2012) addressed these concerns, but used few timepoints to cover a wide developmental period, limiting the conclusions.  In the current study, we utilize an exciting group-based semi-parametric modeling strategy on a large sample of well-characterized children, with many assessment periods across pre-school to school-age. 

Objectives:  To identify distinct developmental trajectories of adaptive behavior, and their predictors, across early-to-late childhood in children with ASD. Although individual domain standard scores and age equivalents were analyzed, for brevity, only the Adaptive Behavior Composite (ABC) results are presented in the abstract.

Methods:  Participants were 104 children with DSM-IV-TR Autistic Disorder (82% male; NVDQ=58.7±17.5, range 19-109; ADOS calibrated severity score=7.6±1.4, range 4-10; 33% minimally verbal) who participated in a longitudinal observational study. The mean age at first assessment was 4.0±1.3 years (range 1.6-7.0), and subjects returned up to six times (mean assessments 3.8±1.1). Data were divided into 10 epochs (6-month periods before 3.5 years, followed by 12-month periods). Proc TRAJ (Jones, Nagin, & Roeder, 2001) for SAS Version 9.3 was used to identify distinct trajectories in VABS-II standard scores across time. The Bayesian Information Criterion (BIC) was used to select the censored-normal model that best fit the data. Model parameters were estimated using Full Information Maximum Likelihood; therefore, missing data are accounted for. Three time-invariant risk factors were entered into the best-fitting model: cognitive ability, language level, and autism symptom severity. 

Results:  Four trajectories best fit the data (Figure 1). All trajectories began with moderately-low to low standard scores (i.e., 70-80), but were characterized as (1) steep decline/very low, (2) moderate decline/very low, (3) steady/moderately low, and (4) low average and rising. Autism severity was not associated with trajectory group membership, but NVDQ and language ability were robust predictors (Table 1). Lower NVDQ was associated with greater likelihood of group membership in the lower-functioning trajectory group in all between-group comparisons. Children who were minimally verbal at initial assessment were at significantly greater risk of belonging to groups 1 and 2 relative to group 3.

Conclusions: Distinct trajectories of overall adaptive behavior in individuals with ASD manifest across childhood. Both NVDQ and minimally verbal status were significant predictors of trajectory; importantly, they distinguished even between similar trajectory groups (i.e., the low but steady versus low and declining). The results of the analyses with the domain scores, which showed differential patterns over time, will be presented at the meeting.