机器学习算法对抑郁症患者自杀企图的识别  

Identification of suicide attempt in patients with depression by machine learning algorithm

作　　者：余涛[1] 刘修燕 许春园 张许来 YU Tao;LIU Xiu-yan;XU Chun-yuan;ZHANG Xu-lai(Affiliated Psychological Hospital of Anhui Medical University,Anhui Mental Health Center,Hefei Fourth People's Hospital,Hefei 230032,China)

机构地区：[1]安徽医科大学附属心理医院,安徽省精神卫生中心,合肥市第四人民医院,230032

出　　处：《临床精神医学杂志》2023年第2期99-102,共4页Journal of Clinical Psychiatry

基　　金：合肥市第四人民医院院内项目(HFSY202102);安徽医科大学校科研基金(2022xkj119)。

摘　　要：目的:很多抑郁症患者自杀企图相关的危险因素已经被确认,但是难以将他们整合成一个模型,用于区分抑郁症患者是否伴有自杀企图。本研究旨在使用机器学习算法结合临床资料,以识别抑郁症患者是否伴有自杀企图。方法:共纳入240例抑郁症患者,按照7∶3比例随机分为训练集和验证集。6种机器学习算法被使用,分别是K最近邻(K-nearest neighbor,KNN)、一般线性模型(general linear model,GLM)、随机森林(random forest,RF)、朴素贝叶斯(naive bayes,NB)、决策树(RPART)和支持向量机(support vector machine,SVM)。在训练集,Logistic回归分析确定的危险因素和10折交叉验证用于模型的构建,验证集以AUC值评估模型效果,选出表现最佳的算法。结果:抑郁症患者自杀企图的发生率为25%。确定低三酰甘油水平和低年龄(≤18岁)是抑郁症患者自杀企图的主要危险因素。GLM是6种机器学习算法中表现最佳的一个,其AUC值达到0.687。结论:机器学习算法能够有效预测抑郁症患者的自杀企图,且GLM表现最佳,有助于防控措施的及时实施,降低抑郁症患者的自杀率。Objective:Many risk factors related to suicide attempts in patients with depression have been identified,but it is difficult to integrate them into a model for distinguishing whether patients with depression have suicide attempts.The purpose of this study is to use machine learning algorithm combined with clinical data to predict whether patients with depression have suicide attempts.Method:A total of 240 patients with depression were included in this study.Patients were randomly divided into training set and verification set according to the ratio of 7∶3.Six machine learning algorithms were used,namely K-nearest neighbor(KNN),general linear model(GLM),random forest(RF),naive bayes(NB),RPART and support vector machine(SVM).In the training set,risk factors identified by the Logistic analysis and 10 fold cross validation were used to construct the model.The effect of the model was evaluated with AUC value during validation set and the best algorithm was selected.Results:The incidence of suicide attempt in patients with depression was 25%.Low triglyceride level and low age(≤18 years)were determined as main risk factors for suicide attempts of patients with depression.GLM was the best one among the six machine learning algorithms,and its AUC reached 0.687.Conclusion:Machine learning can effectively predict the suicide attempt of patients with depression,and GLM has the best performance,which is helpful to the timely implementation of prevention and control measures and reduce the suicide rate of patients with depression.

关 键 词：抑郁症 自杀 机器学习 影响因素 

分 类 号：R749.4[医药卫生—神经病学与精神病学]