Early identification of children with Attention-Deficit/Hyperactivity Disorder (ADHD)

Area of Science:

• Pediatric Health
• Machine Learning in Healthcare
• Developmental Psychology

Background:

• Attention-Deficit/Hyperactivity Disorder (ADHD) symptoms often appear in preschool but are frequently missed, delaying crucial early intervention.
• Early identification of ADHD is vital for effective management and improved long-term outcomes in children.
• Population-level data offers a scalable approach to identify developmental risks like ADHD.

Purpose of the Study:

• To develop and validate machine learning models for the early prediction of ADHD in kindergarten-aged children.
• To assess the utility of combining administrative health data with the Early Development Instrument (EDI) for ADHD detection.
• To identify key predictive factors for ADHD using population-level data.

Main Methods:

• A cohort of 23,494 kindergarten children without prior ADHD diagnosis was analyzed.
• Machine learning models were trained and tested using administrative health data and EDI scores.
• A four-year follow-up period was used to identify incident ADHD cases (1,680 children).

Main Results:

• The best machine learning model achieved an Area Under the Curve (AUC) of 0.811 in predicting ADHD.
• Key predictors included EDI subdomain scores, child's sex, and socioeconomic status.
• The model demonstrated reliable prospective prediction of ADHD using integrated data sources.

Conclusions:

• Machine learning algorithms integrating administrative and EDI surveillance data show promise for early ADHD identification.
• This approach can support timely interventions for children at risk of ADHD.
• Population-level data analysis offers a powerful strategy for proactive child health surveillance.