Categorical Meets Dimensional: A Fuzzy Categorical Conception of Autism Spectrum

Background: Autism spectrum disorder (ASD) is presumed to have a categorical distinction from other disorders and the general population. Much current research, however, concentrates on testing whether ASD is better described by dimensional traits defined over continua. This oversimplifying dichotomy of “categorical vs. dimensional” obscures more finely graded possible alternatives, corresponding to structures that are neither purely categorical nor purely dimensional. In order to give a rigorous account of the nature of ASD, dimensional and categorical aspects of the disorder should be considered in a unified formulation.

Objectives: To demonstrate that a fuzzy categoryconcept can better represent the ASD phenotype. A fuzzy category lacks sharp boundaries and is neither a discrete category nor a perfect continuum, including intermediate cases between those who are clearly affected and those clearly unaffected. In addition, we aim to show that an underlying ASD severity dimension can account for the heterogeneity and the overlapping nature of subcategories.

Methods: Our dataset consisted of 768 participants (678 males, 90 females): 554 children (age: 15.2 ± 3.3 years) with ASD (confirmed by a clinical expert using the ADOS and other assessments), and 214 typically developing children (TDCs) (age: 16.3 ± 3.2 years). A comprehensive phenotypic battery covered domains of social communication, language abilities, executive functioning, general intelligence, anxiety, attention, and hyperactivity. Using taxometric analyses, complex network analysis, and probabilistic inference with the phenotypic data, we defined the sample’s categorical and dimensional characteristics. Finally, we attempted to validate those phenotypic results using brain connectivity analyses with a subset of the sample that had diffusion tensor imaging (DTI) data. 283 participants (239 male and 44 female) had diffusion imaging data (150 ASD and 133 TDC).

Results: The taxometric analyses suggested a categorical structure. Using complex network analysis, we identified communities of participants at multiple neighborhood scales, based on phenotypic relationship between participants. This yielded multiple communities including an intermediate one with a mixture of individuals from both ASD and TDC samples. The identified communities were not stable across different neighborhood scales, indicating heterogeneity and variations inside putative categories. Dynamics of community formation at different scales, revealed clear dimensional effects, with ASD severity regulating participant behavior in forming their communities. Neurobiological analyses, supporting our fuzzy categorical model, resulted in both categorical connectivity differences, and dimensional severity-related alterations.

Conclusions: We provided a computational portrait of ASD phenomenology, as described by phenotypic measures of multiple behavioral domains. Our results confirmed the intricate nature of ASD with categorical and dimensional aspects complementing each other. Even when we tried to identify putative categories in our sample, we saw clear dimensional effects, with ASD severity modulating phenotypic relationship between participants. Our neurobiological results further supported our hypothesis that ASD phenomenology can best be explained by a hybrid model, combining both categorical and dimensional structures. It should be noted, however, that our results are specific to our sample, which include those who met criteria for ASD via gold standard assessments and those who qualify as TDCs. Future work should include unselected clinical samples.