A Random Forest-Based Risk Prediction Model for Non-Response to Methylphenidate in Children with Attention Deficit Hyperactivity Disorder.

Abstract

Objective: To develop and validate a prediction model based on the random forest algorithm to assess the risk of non-response to methylphenidate (MPH) in children with attention deficit hyperactivity disorder (ADHD), thereby providing decision support for individualized clinical treatment.

Methods: A total of 150 children with ADHD who received MPH treatment were prospectively and consecutively enrolled. Based on changes in the Swanson, Nolan, and Pelham Rating Scale, Fourth Edition (SNAP-IV) scores after 3 months of treatment, patients were classified into a treatment response group (n = 116) and a non-response group (n = 34). Differences in clinical characteristics, pre-treatment clinical assessment scales, and laboratory parameters were compared. A random forest algorithm was used to construct the prediction model. Feature importance was evaluated based on the decrease in node impurity.

Results: Serum 25-hydroxyvitamin D [25(OH)D], cortisol, S100β protein, brain-derived neurotrophic factor (BDNF), and urinary catecholamines were lower in the treatment non-response group. The combined subtype of ADHD, higher SNAP-IV total score, lower 25(OH)D levels, and lower dopamine levels were independent risk factors for non-response. In the random forest model, SNAP-IV score had the highest feature importance. The model, incorporating clinical characteristics, pre-treatment clinical assessment scales, and laboratory indicators, demonstrated excellent predictive performance in the test set (AUC=0.883; 95% confidence interval: 0.811-0.956) and an overall accuracy of 86.67%.

Conclusion: The random forest-based prediction model can accurately identify children with ADHD who are unlikely to respond to MPH treatment.