Digital phenotyping captures autism-associated behaviors in preschool- and school-age autistic children with and without co-occurring ADHD

Abstract

Abstract There is a need for scalable, objective assessment tools to quantify autism-related behaviors in preschool- and school-age children. A significant challenge is the heterogeneous presentation of autism, driven in part by co-occurring conditions such as Attention-Deficit/Hyperactivity Disorder (ADHD). Tools intended for autism must therefore be tested in samples that include ADHD and other comorbidities, not only in autism-versus-neurotypical comparisons. SenseToKnow, a digital phenotyping app, quantifies autism-related behaviors using computer vision, tactile sensors, and machine learning, distinguishing autistic and neurotypical toddlers. We administered SenseToKnow to 183 children aged 40–100 months (3.3–8.3 years): 41 neurotypical, 48 ADHD, 53 autism, and 41 co-occurring autism and ADHD. Two complementary analyses converged. In age-adjusted group comparisons, autistic children, with and without ADHD, exhibited different SenseToKnow features compared to neurotypical and ADHD children, while autistic children with and without ADHD did not differ; children with ADHD alone differed from neurotypical children, particularly during nonsocial stimuli. Furthermore, in regression modeling, autism status was associated with 21 of 23 SenseToKnow features and ADHD status with none. Across both analysis, SenseToKnow features tracked autism status rather than ADHD status, even when the two co-occurred. These results demonstrate that SenseToKnow captures autism-associated behaviors even in the presence of co-occurring ADHD.