Background: Autistic children and adolescents often demonstrate motor dysfunctions, prominently including toe walking (TW). Prior investigations have identified persistent TW (i.e., TW that fails to resolve autonomously by the ages of 3-7 years) as being correlated with increased cognitive deficits, delays in language development, and orthopaedic issues, whereas the evidence pertaining to atypical sensory processing remains less uniform. These factors are characterised by non-linearity, complexity, and fuzzy interaction, necessitating a machine learning approach.

Objectives: The goal of this study is to describe the overall schema of natural relationships between clinical variables (i.e., four quadrants and six sensory systems) of Sensory Profile 2 (SP-2) scores in a sample of autistic children and adolescents with and without TW, using fourth–generation artificial neural networks.

Methods: The caregivers of 40 children and adolescents with ASD (12 females, 28 males; mean age = 8.5 years, SD = 2.54), 16 with TW and 24 without TW, filled out the SP-2. The CARS-2 scale was used to classify the severity of autism. To illustrate the most important connections among the variables, a non-supervised artificial neural network called Auto-CM, coupled with a Minimum Spanning Tree filter, developed a Semantic Connectivity Map of the variables in the study. According to SP-2 user manual, three classes of variables pertaining to SP-2 were formed as data preprocessing: a class with values in the range of “just like the majority of cases” (Typical), a class with values in the range of “more or much more” (More), and a class with values in the range of “less or much less” (Less), with an appropriate transformation.

Results: The semantic connectivity map developed by the Auto-CM system clearly shows a high level of complexity with many variables acting as hubs with three or more links. The map revealed a clear separation between the presence/absence of TW. The TW-present node was directly linked with the cluster of SP-2 subscales “More” values through SP-2 sensitivity. The hierarchy of SP-2 subscales with “more” values according to degrees of separation was as follows: sensitivity (first level), auditory and registration (second level), avoiding, touch and seeking (third level), body position and movement (fourth level), and visual (fifth level). The TW absent node was directly linked with the cluster of SP-2 “Typical” subscales through SP-2 body position. These connections remained unchanged after progressive addition of other variable domains, like motor domain (Developmental Coordination Disorder Questionnaire scores) and autism severity, underlying an elective association between TW and showing "more" and/or "much more" sensitivity than neurotypical persons (Figure 1). The only SP-2 subscale not showing this elective association was “oral sensitivity”.

Conclusions: These results highlight the strong dependence of TW on atypical sensory processing in which sensitivity, auditory and registration/bystander play a major role. Further research on a larger sample is warranted to confirm the results.

Notes: 

(1)School of Medicine and Surgery, University of Milano-Bicocca, Milano, Milano, Italy, (2)PhD Program in Neuroscience, University of Milano-Bicocca, Milano, Milano, Italy, (3)Autism Research Unit, Villa Santa Maria SCS, Tavernerio, Como, Italy