Prediction of Treatment Response and Adverse Events in Pediatric Attention-Deficit/Hyperactivity Disorder Treated

Yang Li1, Zilong Ma2, Walter Falchi2

  • 1The First Affiliated Hospital of Heilongjiang University of Chinese Medicine, Harbin, China.

Insights

Machine learning models accurately predict treatment response and adverse events in children with attention-deficit/hyperactivity disorder (ADHD) receiving acupuncture. These models aid in personalized risk assessment for ADHD treatment.

Area of Science:

  • Pediatric Neurology
  • Integrative Medicine
  • Computational Psychiatry

Background:

  • Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder in children.
  • Conventional first-line treatments for ADHD may have limitations in efficacy and side effects.
  • Acupuncture is explored as an adjunctive therapy to conventional care for pediatric ADHD.

Purpose of the Study:

  • To develop and externally validate machine learning (ML) models for predicting treatment response in children with ADHD.
  • To develop and externally validate ML models for predicting adverse events in children with ADHD.
  • To assess the clinical utility and interpretability of these ML models.

Main Methods:

  • A multicentre retrospective study involving 809 children (aged 6-12 years) with ADHD.
  • Development and evaluation of ten ML models using demographic, clinical, behavioral, and laboratory data.
  • Models were validated internally and in two external cohorts using 10-fold cross-validation.

Main Results:

  • An extreme gradient boosting model showed the best performance for predicting treatment response (AUROC 0.78-0.87).
  • Moderate discriminative performance was observed for predicting adverse events (external AUROCs 0.73-0.85).
  • Interpretability analyses indicated that multiple clinical features contribute to model predictions.

Conclusions:

  • ML models using routinely collected data can reliably predict treatment response and adverse events in pediatric ADHD patients receiving acupuncture.
  • These models may facilitate individualized benefit-risk assessments in integrative care settings.
  • Prospective studies are needed to confirm the clinical impact of these ML models.
Abstract