Machine learning-based prediction of intellectual disability in children with autism spectrum disorder: using behavioral observation techniques.

Autism Spectrum Disorder (ASD) is a label for a collection of symptoms that begin in early childhood and involve major but variable impairments in social communication and stereotypical or repetitive behaviors []. ASD prevalence has dramatically increased from approximately 1 in 10,000 in 1980 to 323 per 10,000 in 2022 in the United States [, ]. Average lifetime care costs are estimated at $3.6 million per individual; however, individuals severely affected by the disorder can require much more expensive care []. Early behavioral intervention can reduce the core symptoms of ASD and improve outcomes in later life [, ]. Behavioral interventions are most effective beginning in early infancy, preferably in the first 24 months of life []; however, the average age of first diagnostic assessment interview in the United States (US) is 47 months []. Thus, there exists an urgent need to develop a non-invasive screening tool to identify those at high risk for ASD to facilitate earlier diagnosis and treatment [, ]. Identifying children with ASD as young as possible and treating them can improve long-term outcomes, enhance quality of life, and potentially save hundreds of billions of dollars per year []. ASD is thought to result from a complex interplay of genetic and environmental factors []. Different factors are important for different people, with about 10% of cases involving major genetic disorders such as Fragile X, and about 90% of cases being of unknown origin. Furthermore, stratifying the heterogeneous ASD population into distinct, well-defined phenotypes will likely improve treatment, as targeted therapeutic plans can be developed for each phenotype.