Machine Learning Based Early Diagnosis of ADHD with SHAP Value Interpretation: A Retrospective Observational Study

• Published in: Neuropsychiatric disease and treatment Vol. 21; no. Issue 1; pp. 1075 - 1090
• Main Authors: Zhang, Xinyu, Xiao, Xue, Luo, Yufan, Xiao, Wei, Cao, Yingsi, Chang, Yuanjin, Wu, Dongqin, Xu, Hua, Zhao, Jinlin, Deng, Xianhui, Jiang, Yuanying, Xie, Ruijin, Liu, Yueying
• Format: Journal Article
• Language: English
• Published: New Zealand Dove Medical Press Limited 01.01.2025; Taylor & Francis Ltd; Dove; Dove Medical Press
• Subjects: ADHD; Algorithms; Attention deficit hyperactivity disorder; Biomarkers; Blood levels; Blood sugar; Computer-aided medical diagnosis; Diagnosis; Electrolytes; Erythrocytes; Feature selection; Guardians; Hospitals; Hyperactivity; Impulsive behavior; Impulsivity; Learning algorithms; Leukocytes (eosinophilic); Machine learning; Mental disorders; Methods; Neurodevelopmental disorders; Observational studies; Original Research; Pediatrics; SHAP methods; Vitamin D3; China; Taiwan; ADHD; diagnosis; SHAP methods; machine learning; biomarkers
• ISSN: 1178-2021; 1176-6328; 1178-2021
• DOI: 10.2147/NDT.S519492

Attention-Deficit/Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder in children, characterized by inattention, hyperactivity, and impulsivity. Current diagnostic methods for ADHD rely primarily on behavioral assessments, which can be challenging due to symptom overlap with other psychiatric disorders and significant inter-individual variability. Developing potential early diagnostic methods for ADHD is imperative to mitigate the risk of misdiagnosis and enhance the evaluation of treatment efficacy. The study was conducted at the Department of Pediatrics, Affiliated Hospital of Jiangnan University, from November 2022 to January 2024. Clinical data, including complete blood count, liver and kidney function tests, blood glucose levels, serum electrolyte tests, and serum 25-dihydroxyvitamin D3 levels, were collected. Feature selection and model construction were performed using various machine learning algorithms. Our results indicated that the Gradient Boosting Machine algorithm is the optimal model. Our machine learning analyses suggest that the Gradient Boosting Machine (GBM) model may be the optimal choice, highlighting blood beta-2 microglobulin levels, red blood cell distribution width, 25-dihydroxyvitamin D3, and the percentage of eosinophils as key predictors of ADHD risk, thereby aiding early diagnosis. Further large-scale studies are warranted to validate these findings and explore the underlying mechanisms.