ISSN 1674-3865  CN 21-1569/R
主管：国家卫生健康委员会
主办：中国医师协会
　　　辽宁省基础医学研究所
　　　辽宁中医药大学附属医院
中国中西医结合儿科学

中国中西医结合儿科学 ›› 2024, Vol. 16 ›› Issue (2): 130-136.

• 临床研究 • 上一篇    下一篇

基于机器学习的注意力缺陷多动障碍风险预测研究

  

  • 出版日期:2024-04-25 上线日期:2024-04-25

Risk prediction of attention deficit hyperactivity disorder based on machine learning

  • Published:2024-04-25 Online:2024-04-25

PDF (PC)

17

摘要： 目的 探讨基于机器学习算法对儿童注意力缺陷多动障碍(ADHD)预测的可行性。方法 回顾性分析我院于2022年11月至2023年8月儿科门诊就诊患者358例，其中ADHD患儿119例，非ADHD患儿239例，以人口学基本信息、儿童个人生活情况、母亲孕期情况、家庭生活情况及遗传因素等31个变量作为危险因素，采用单因素分析筛选出具有明显差异的变量，然后分别建立决策树模型、随机森林模型、自适应提升算法及K近邻算法模型，采用受试者工作特征(ROC)曲线的面积(AUC)、特异度、准确性、F1分数及ROC曲线等进行模型预测效能评估。结果 4种机器学习算法建立的ADHD的预测模型以随机森林算法最优，其AUC为0.955,特异度、准确性、F1分值分别为0.903、0.898、0.853;同时，根据随机森林模型筛选出的前五位特征变量为：教育方式、情绪稳定情况、每日看电子产品时长、学习困难情况、近期反复呼吸道感染。结论 初步构建出基于机器学习算法建立儿童ADHD的预测模型，该模型对ADHD有良好的预测能力。

关键词: 注意力缺陷多动障碍, 危险因素, 机器学习, 预测模型, 儿童

Abstract: ObjectiveTo explore the feasibility of predicting attention deficit hyperactivity disorder(ADHD) in children based on machine learning algorithm.MethodsA total of 358 patients treated in the pediatric outpatient department of our hospital from November 2022 to August 2023 were retrospectively analyzed,and 119 patients were finally included in the ADHD group and 239 patients in the non-ADHD group.Totally 31 variables,including basic demographic information,children's personal life situation,mother's pregnancy situation,family life situation and genetic factors,were taken as risk factors.Single factor analysis was used to select variables with obvious differences,and then the decision tree(DT) model,random forest(RF) model,adaptive enhancement algorithm(Adaboost) and K-nearest neighbor algorithm(KNN) models were established respectively.AUC,specificity,accuracy,F1 score and ROC curve were used to evaluate the model prediction efficiency.ResultsRandom forest algorithm was the best predictive model for ADHD,with AUC being 0.955,and specificity,accuracy and F1 scores being 0.903,0.898 and 0.853,respectively.Meanwhile,the top five characteristic variables screened according to the random forest model were:education style,emotional stability,daily time spent playing with electronic products,learning difficulties,and recent recurrent respiratory infections.ConclusionA prediction model of child ADHD based on machine learning algorithm is established,which has good prediction ability for ADHD.

Key words:

Attention deficit hyperactivity disorder, Risk factors, Machine learning, Prediction model, Children