Feature Selection and Prediction of Pediatric Tuina in Attention Deficit/Hyperactivity Disorder Management: A Machine Learning Approach Based on Parent-Reported Children's Constitution

Area of Science:

• Integrative medicine
• Pediatric neurology
• Computational biology

Background:

• Attention Deficit/Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental disorder in children.
• Pediatric tuina, a Traditional Chinese Medicine (TCM) therapy, shows promise for managing ADHD symptoms.
• Integrating machine learning (ML) can personalize tuina treatments for better parent-led administration.

Purpose of the Study:

• To develop an ML-based model for predicting individualized pediatric tuina treatments for ADHD.
• To identify key features for TCM pattern identification and personalized ADHD management.
• To evaluate the efficacy of ML models in guiding tailored tuina interventions.

Main Methods:

• An ML model analyzed parent-reported constitutional data from 1005 children with ADHD.
• Feature selection was performed using models like SVM, LR, MLP, and RF.
• The study focused on identifying relevant features for TCM pattern diagnosis and treatment personalization.

Main Results:

• ML models demonstrated strong predictive capabilities for personalized tuina treatments.
• The Multilayer Perceptron (MLP) model achieved the highest Area Under the Curve (AUC) of 0.90 and accuracy (ACC) of 0.74.
• Seven key features were consistently selected, enabling targeted pediatric tuina applications.

Conclusions:

• An ML-driven approach was successfully developed to enhance ADHD management through personalized pediatric tuina.
• The study identified seven critical features for TCM pattern identification, guiding individualized treatment strategies.
• ML, particularly MLP, significantly improved the prediction of effective tuina interventions for pediatric ADHD.