Evaluating Validity of the Autism Observation Scale for Infants (AOSI) Among Infants Identified with Possible Autism By Community Care Professionals

Background: The Autism Observation Scale for Infants (AOSI; Bryson et al., 2008) is a brief, play-based assessment developed to quantify the behavioural manifestation of autism among 6-18-month-olds. From the first birthday – though not before 12-months – AOSI Total scores exceeding a threshold of 9 have been associated with likely later diagnosis (e.g., Brian et al., 2006; Bryson et al., 2008). Developers note the AOSI is not yet validated for clinical use (Bryson & Zwaigenbaum, 2014). Further, only one validation study exists beyond the development group (Gammer et al., 2015). The AOSI has also been included predominantly within prospective ‘high-risk sibling’ studies, and there is growing recognition of the need to investigate whether inferences from these generalise to the broader population of infants developing autism (e.g., Sacrey et al., 2017).

Objectives: To conduct an independent evaluation of validity of the AOSI in a community-referred sample.

Methods: Infants aged 9-16 months (n=103; 68% boys) were referred due to showing early signs of autism according to the Social Attention and Communication Surveillance (SACS) protocol (Barbaro & Dissanayake, 2010). Specifically, all infants showed ≥3 (of 5) key markers on the SACS 12-month checklist (i.e., absent/atypical eye-contact, response to name, imitation, pointing, other gestures). Infants attended two assessments, 6-months apart. We administered the AOSI, Mullen Scales of Early Learning (MSEL) and Vineland Adaptive Behavior Scales (VABS) at Time 1 (T1) and Time 2 (T2), with the Autism Diagnostic Observation Schedule–Toddler Module (ADOS-T) also administered at the latter. AOSI assessments were coded live and from video, with 20% of tapes double-coded.

Results: Very high agreement was observed for live vs. video scoring of T1 AOSIs (within-rater live/video agreement on 100% assessments, r=.89; inter-rater agreement on 20% tapes, r=.79). At both T1 and T2, AOSI Total scores approximated a normal distribution, centred on the previously-reported threshold score of 9 (T1: M=9.44, SD=4.05; T2: M=9.32, SD=4.68). At T1, AOSI scores were significantly – though only weakly – associated with concurrent MSEL (r= -.23, p=.017) and VABS totals (r= -.23, p=.023), but were elevated among children with ≥4+ (vs. only 3) SACS markers at referral (F[2,102]=6.73, p=.002). At T2, AOSI scores were moderately associated with MSEL (r= -.36, p<.001) and VABS totals (r= -.43, p<.001), and more strongly associated with ADOS-T Algorithm totals (r=.627, p<.001). AOSI scores showed only moderate T1-T2 association/stability (r=.441, p<.001) and T1 AOSI scores only moderately predicted T2 ADOS-T totals (r=.412, p<.001).

Conclusions: The AOSI is reliably scored and shows some validity for quantifying autism behaviours in a community-referred sample of infants. We observed an approximately normal distribution of AOSI scores in our sample – centred on the previously suggested threshold for likely later diagnosis – but with only moderate/weak within-participant stability over time. Like ‘high-risk siblings’, community-referred infants with early signs of autism may experience variable trajectories of early development. The high-level reliability of video (vs. standard/live) ratings supports the AOSI as a viable measure for inclusion in protocols necessitating blinded evaluation; for example, as a primary outcome measure for pre-emptive intervention trials.