Using Artificial Intelligence to Identify Factors Associated with Autism Spectrum Disorder in Adolescents with Cerebral Palsy.

Abstract

Autism spectrum disorder (ASD) is common in adolescents with cerebral palsy (CP) and there is a lack of studies applying artificial intelligence to investigate this field and this population in particular. The aim of this study is to develop and test a predictive learning model to identify factors associated with ASD in adolescents with CP. This was a multicenter controlled cohort study of 102 adolescents with CP (61 males, 41 females; mean age\u2009±\u2009SD [standard deviation]\u2009=\u200916.6\u2009±\u20091.2 years; range: 12-18 years). Data on etiology, diagnosis, spasticity, epilepsy, clinical history, communication abilities, behaviors, intellectual disability, motor skills, and eating and drinking abilities were collected between 2005 and 2015. Statistical analysis included Fisher s exact test and multiple logistic regressions to identify factors associated with ASD. A predictive learning model was implemented to identify factors associated with ASD. The guidelines of the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement were followed. Type of spasticity (hemiplegia\u2009>\u2009diplegia\u2009>\u2009tri/quadriplegia; OR [odds ratio]\u2009=\u20091.76, SE [standard error]\u2009=\u20090.2785, p\u2009=\u20090.04), communication disorders (OR\u2009=\u20097.442, SE\u2009=\u20090.59, p\u2009<\u20090.001), intellectual disability (OR\u2009=\u20092.27, SE\u2009=\u20090.43, p\u2009=\u20090.05), feeding abilities (OR\u2009=\u20090.35, SE\u2009=\u20090.35, p\u2009=\u20090.002), and motor function (OR\u2009=\u20090.59, SE\u2009=\u20090.22, p\u2009=\u20090.01) were significantly associated with ASD. The best average prediction model score for accuracy, specificity, and sensitivity was 75%. Motor skills, feeding abilities, type of spasticity, intellectual disability, and communication disorders were associated with ASD. The prediction model was able to adequately identify adolescents at risk of ASD.