Development and validation of a machine learning‐based tool to predict autism among children

• Published in: Autism research 2023-05, Vol.16 (5), p.941-952
• Main Authors: Betts, Kim Steven, Chai, Kevin, Kisely, Steve, Alati, Rosa
• Format: Article
• Language: English
• Subjects: administrative data linkage; Analgesia; Apgar score; Australia; Autism; Autism Spectrum Disorder - diagnosis; Autistic Disorder - diagnosis; Child; Data collection; Disorders; Female; Humans; Infant; Infant, Newborn; Learning algorithms; Machine Learning; Maternal Age; Mental health; Offspring; Prediction models; Predictions; predictive models; Pregnancy; Risk factors; Tobacco
• ISSN: 1939-3792; 1939-3806
• DOI: 10.1002/aur.2912

Autism is a lifelong condition for which intervention must occur as early as possible to improve social functioning. Thus, there is great interest in improving our ability to diagnose autism as early as possible. We take a novel approach to this challenge by combining machine learning with maternal and infant health administrative data to construct a prediction model capable of predicting autism disorder (defined as ICD10 84.0) in the general population. The sample included all mother–offspring pairs from the Australian state of New South Wales (NSW) between January 2003 and December 2005 (n = 262,650 offspring), linked across three health administrative data sets including the NSW perinatal data collection (PDC); the NSW admitted patient data collection (APDC) and the NSW mental health ambulatory data collection (MHADC). Our most successful model was able to predict autism disorder with an area under the receiver operating curve of 0.73, with the strongest risk factors for diagnoses found to include offspring gender, maternal age at birth, delivery analgesia, maternal prenatal tobacco disorders, and low 5‐min APGAR score. Our findings indicate that the combination of machine learning and routinely collected admin data, with further refinement and increased accuracy than achieved by us, may play a role in the early detection of autism disorders. Lay Summary Increasing evidence suggests that early intervention programs can improve social functioning outcomes among children with autism, with the best results obtained by starting as early as possible. We aimed to investigate whether machine learning and routinely collected health administrative data could be utilized to predict infants at risk of developing autism disorder by age 14. Our results indicate that with further refinement, this approach could be useful in identifying infants at age 12 months who would likely benefit from early intervention programs.