Identifying Cross-Domain Cognitive Subtypes Among Children, Adolescents and Adults with Autism Spectrum Disorders
Objectives: 1) To identify cross-domain cognitive profiles in a large sample of individuals with ASD who were assessed as part of the EU-AIMS Longitudinal European Autism Project (LEAP); and 2) To investigate whether different cognitive subtypes differ in their clinical symptoms.
Methods: Participants completed a battery of cognitive tests tapping explicit and spontaneous theory of mind (a continuous false belief task and the Animated Shapes task), social attention (% gaze time on faces), spatial working memory (SWM), probabilistic reversal learning (PRL), and WCC (Block Design). Across tasks, 349-435 individuals with ASD and 251-297 individuals with typical development (TD) or mild intellectual disabilities aged 6-30 years were included. First, we tested case-control differences on each task. Second, we estimated developmental growth curves by generating normative models for each TD participant’s [task score] relative to their age using Gaussian Processes for Machine Learning. For each participant with ASD we then quantified the deviation from the normative model. Finally, to identify cognitive subtypes, we used hierarchical clustering based on each individual’s deviations from the means across all tasks.
Results: We found significant mean group differences in social attention (p=.002, d=.26) PRL (p<.0001, d=.35) and SWM (p<.0001, d=.43). However, normative modelling revealed that on the SWM task 66% of participants with ASD performed within +/-1 SD of the age-expected TD means or above; 19.7% fell between 1-2 SDs and 12.8% below 2 SDs. These patterns were very similar for the PRL and social attention measures (Figure 1). Hierarchical clustering (including only participants who completed all tasks) revealed 6 distinct clusters (Figure 2), which partly differed in their symptom presentation. For example, Cluster 6 (impaired ToM+ “intact” EF+ WCC) had on average significantly fewer repetitive behaviors than Cluster 3 (impaired social attention+ impaired SWM) and higher levels of adaptive behaviour than Clusters 3 and 2 (impaired reversal learning + other impairments), perhaps reflecting their ability to recruit ‘intact’ EF skills as a compensatory mechanism.
Conclusions: Using a battery of ‘classic’ ASD-related cognitive tasks, we show distinct cognitive profiles among people with ASD, with partly differing clinical symptoms. These results point to the value of stratification approaches to reduce heterogeneity, to refine both aetiology and intervention.