基于随机森林算法和多因素logistic回归分析模型的孕期抑郁影响因素  

Exploring the influencing factors of antenatal depression based on random forest algorithm and multivariate logistic regression model

作　　者：邓淦元 杨中婷 杜兴梅 邓春燕 宋婷 李瑞雨 何杰宇 邓睿 黄源 陈莹 DENG Ganyuan;YANG Zhongting;DU Xingmei;DENG Chunyan;SONG Ting;LI Ruiyu;HE Jieyu;DENG Rui;HUANG Yuan;CHEN Ying(School of Public Health,Kunming Medical University,Kunming 650500,China)

机构地区：[1]昆明医科大学公共卫生学院,昆明650500

出　　处：《中华疾病控制杂志》2023年第9期1003-1009,共7页Chinese Journal of Disease Control & Prevention

基　　金：国家自然科学基金资助项目(72264020);中华医学基金会资助项目(CMB#19-338);云南省中青年学术和技术带头人后备人才项目(202305AC160046);健康与危害量化测评博士生导师团队建设项目(2022)。

摘　　要：目的分析孕期抑郁检出率探讨可能的影响因素,识别高危人群为预防抑郁症提供科学依据。方法对云南省某县2022年5月处于孕期的女性人群采用爱丁堡产后抑郁症量表(Edinburgh Postnatal Depression Scale,EPDS)进行抑郁筛查,利用随机森林算法对影响因素进行重要性排序,用滑动窗口序贯向前选择法降维,将重要性评分最高且平均袋外估算误差率最小的影响因素纳入多因素logistic回归分析模型,估计影响因素的作用方向及效应值。结果732名孕妇接受问卷筛查,孕期抑郁检出率为13.8%(101/732)。随机森林算法分析显示,变量数为7时平均袋外估算误差率最小。将重要性排名前7的影响因素纳入多因素logistic回归分析模型,结果显示:孕期焦虑、既往不良情绪史、生孩子的经济担忧、家人对孩子有性别期盼是孕期抑郁的危险因素,第2次怀孕、高社会支持水平是孕期抑郁的保护因素,孕妇自评健康状况不满意与孕期抑郁无关。结论孕期有焦虑的女性很可能伴有抑郁,对生育费用、孩子性别的担忧和既往的不良情绪史会增加孕期抑郁的风险,足够的社会支持和既往的孕育史可以减少孕期抑郁发生的风险。建议加强孕期抑郁筛查,早期识别孕期抑郁症的高危人群。Objective To analyze the prevalence of antenatal depression,to explore the possible influencing factors of antenatal depression,and and to recognize high-risk groups,thereby contributing to prevention strategies.Methods In May 2022,pregnant women in a county of Yunnan Province were surveyed and antenatal depression were screened by Edinburgh Postpartum Depression Scale(EPDS).Random forest algorithm was adopted to rank the importance of the influencing factors of antenatal depression.The sliding window sequential forward selection method was employed to decide the number of influencing factors.The influencing factors with the highest importance score and the lowest out-of-bag error were incorporated into the Multivariate logistic regression model to estimate the effect size.Results A total of 732 pregnant women were screened by questionnaire,the prevalence of antenatal depression was 13.8%(101/732).The out-of-bag error was lowest when the number of influencing factors was 7 based on random forest algorithm and sliding window sequential forward selection.We included the top 7 influencing factors in the logistic regression model,and the results showed that antenatal anxious,history of negative emotions,financial concerns of having children,and baby gender expectation were the risk factors for antenatal depression.In contrast,the second pregnancy,high social support were the protective factors.Self-reported poor health was found unrelated to antenatal depression.Conclusions Women with anxiety during pregnancy are more likely to develop depression.The concerns of financial costs,baby gender and bad emotions experience increase the risk of triggering depression during pregnancy,while the adequate social support and pregnancy experience reduce the risk.It is recommended that high-risk group screening should be given to women during the pregnancy to prevent antenatal depression.

关 键 词：孕期抑郁 随机森林算法 多因素logistic回归分析模型 影响因素 

分 类 号：R181.2[医药卫生—流行病学] R749.4[医药卫生—公共卫生与预防医学]