Background: Currently, there is a need for approaches to understand and manage the multidimensional autism spectrum and quantify its heterogeneity. The diagnosis is based on behaviors observed in two key dimensions, social communication and repetitive, restricted behaviors, alongside the identification of required support levels. However, it is now recognized that additional modifiers, such as language abilities, IQ, and comorbidities, are essential for a more comprehensive assessment of the complex clinical presentations and clinical trajectories in autistic individuals. Different approaches have been used to identify autism subgroups based on the genetic and clinical heterogeneity, recognizing the importance of autistic behaviors and the assessment of modifiers. While valuable, these methods are limited in their ability to evaluate a specific individual in relation to a normative reference sample of autistic individuals. A quantitative score based on axes of phenotypic variability could be useful to compare individuals, evaluate the homogeneity of subgroups, and follow trajectories of an individual or a specific group. Here we propose an approach by (i) combining measures of phenotype variability that contribute to clinical presentation and could impact different trajectories in autistic persons and (ii) using it with normative modeling to assess the clinical heterogeneity of a specific individual.
Methods: Using phenotypic data available in a comprehensive reference sample, the Simons Simplex Collection (n = 2744 individuals), we performed principal component analysis (PCA) to find components of phenotypic variability. Features that contribute to clinical heterogeneity and could impact trajectories in autistic people were assessed by the Autism Diagnostic Interview-Revised (ADI-R), Vineland Adaptive Behavior Scales (VABS) and the Child Behavior Checklist (CBCL). Cognitive assessment was estimated by the Total Intelligence Quotient (IQ).
Results: Three PCs embedded 72% of the normative sample variance. PCA-projected dimensions supported normative modeling where a multivariate normal distribution was used to calculate percentiles. A Multidimensional General Functionality Score (MGFS) to evaluate new prospective single subjects was developed based on percentiles.
Conclusions: Our approach proposes a basis for comparing individuals, or one individual at two or more times and evaluating homogeneity in phenotypic clinical presentation and possibly guides research sample selection for clinical trials.
Keywords: autism spectrum disorder; normative modeling; phenotypic heterogeneity; principal component analysis.