Assessment of Social Communication in Infants At High Risk for Autism Spectrum Disorders: Inclusion of Naturalistic Behavior Samples

Background: Deficits in social communication behaviors has been documented in infants and older children with ASD (e.g., Wetherby et al., 2004), and are a core feature of the disorder (DSM-IV-TR; American Psychiatric Association, 2000). Infant siblings of children with ASD are at heightened biological risk of developing ASD themselves (HR; e.g., Zwaigenbaum et al., 2009; Ozonoff et al., 2011). While symptoms of ASD are reliably measured by 18 months of age (e.g., Lord, 1995; Stone et al., 1999), a diagnosis is rarely given before the age of 2 or 3 (e.g., Turner et al.2006). Although some argue that assessment of HR infants include natural communication samples (e.g.,Tager-Flusberg et al., 2009), very little work has followed this recommendation.  

Objectives: To examine whether communicative behavior samples collected in HR infants’ natural environments, in combination with commonly utilized standardized assessments, can better inform prediction of diagnostic outcome. 

Methods: Fifty HR infants (44% male) were observed at home with a primary caregiver at 8, 10, 12, 14, and 18 months of age. Frequencies of infant-initiated gestures, words, non-word vocalizations, eye contact, and smiles were coded from 25-minute videotaped observations of everyday activities and parent-child toy play. At each visit, caregivers completed the MacArthur-Bates Communicative Development Inventory (CDI; Fenson et al., 2007), and infants were administered the Early Social Communication Scales (ESCS; Mundy et al., 2003). All HR infants received a diagnostic evaluation at 36 months (i.e., ADOS and clinical judgment using DSM-IV-TR criteria); nine HR infants were diagnosed with ASD (HR-ASD). Thirteen HR infants met criteria for a language delay (Language Delay; HR-LD; Heilmann et al., 2005). The remaining 28 HR infants did not meet ASD or language delay criteria (No Diagnosis; HR-ND). 

Results: First, a discriminant function analysis (DFA) was run using only data gathered from the ESCS and CDI. The overall Wilks’ lambda was significant, (Ʌ = 0.50, χ² (18, N = 46) = 29.61, p = .041), indicating that overall the predictors differentiated among the three groups. However, only 73.9% of the cases were correctly classified based on the ESCS and CDI scores (sensitivity = 100%, specificity = 67.6%). Next, to determine whether or not diagnostic prediction is improved by adding data gathered through natural communication samples, another DFA was conducted, adding key variables from the naturalistic observation. As before, results were significant, Ʌ = 0.07, χ² (48, N = 46) = 81.33, p= .002. However, the percentage of children correctly classified as HR-ND, HR-ASD, or HR-LD, increased from 73.9% to 93.3%. All of the HR children diagnosed with ASD at 36 months were correctly identified by the predictors in the model, yielding 100% sensitivity. Importantly, specificity (91.7%) was much improved over the previous model. 

Conclusions: Overall, results demonstrated that behavior samples gathered from a naturalistic play setting improved the ability to predict whether HR infants were later classified as ASD or LD. Findings support the notion that multi-method sampling procedures that incorporate structured evaluation, parent report, and measures derived from naturalistic interactions may improve screening and diagnosis of ASD.