基于随机森林模型的儿童青少年近视影响因素分析  

Analysis of influencing factors of myopia in children and adolescents based on random forest model

作　　者：林米乐 彭慧[2] 杨飞飞 薛俊磊 张颜菲 于宏杰[2] 马飞飞[2] 彭谦[2] 王颖[3] 史慧静[3] LIN Mi-le;PENG Hui;YANG Fei-fei;XUE Jun-lei;ZHANG Yan-fei;YU Hong-jie;MA Fei-fei;PENG Qian;WANG Ying;SHI Hui-jing(Peking Union Medical College,Institute of Medical Information,Chinese Academy of Medical Sciences,Beijing 100020,China;不详)

机构地区：[1]北京协和医学院中国医学科学院医学信息研究所,北京100020 [2]上海市嘉定区疾病预防控制中心,上海201800 [3]复旦大学公共卫生学院,上海200023

出　　处：《现代预防医学》2024年第21期3905-3910,共6页Modern Preventive Medicine

基　　金：嘉定区公共卫生优秀人才培养计划-学科后备带头人项目;“沪眼星光”儿童青少年近视防控人员培养项目(HYXG-QJ04);嘉定区卫生健康委项目(2022-QN-08);复旦大学公共卫生学院-嘉定区卫生健康委公共卫生高质量发展重点专项(GWGZLXK-2023-04)。

摘　　要：目的分析儿童青少年近视影响因素,为精准预防近视提供参考。方法采用分层整群随机抽样的方法,于2021年9—10月选取上海市嘉定区6所学校1660名9~18岁学生,进行视力检测和问卷调查,利用随机森林模型对近视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1660名儿童青少年近视率71.99%。随机森林模型显示,变量数为6时平均袋外估算误差率最小。将其纳入多因素logistic回归分析模型,结果显示:学段增加(初中:OR=5.191,95%CI:3.738~7.209;高中:OR=6.989,95%CI:4.309~11.338)、父母近视(单方近视:OR=3.669,95%CI:2.723~4.943;双方近视:OR=5.231,95%CI:3.510~7.795)、读写时长增加(<1h:OR=12.192,95%CI:3.373~44.066;1~2h:OR=11.985,95%CI:3.364~42.704;≥2h:OR=10.547,95%CI:2.957~37.622)、参加课外学习班(<6岁:OR=3.114,95%CI:2.046~4.740;≥6岁:OR=1.817,,95%CI:1.351~2.444)、天黑后在家读写仅使用屋顶灯(OR=2.921,95%CI:2.009~4.248)是近视的危险因素,睡眠时长≥8h(OR=0.383,95%CI:0.264~0.557)是近视的保护因素。结论儿童青少年近视率较高,近视防控应根据影响因素采取针对性措施。Objective To analyze the influencing factors of myopia in children and adolescents,providing references for precise myopia prevention.Methods A stratified cluster random sampling method was employed to select 1660 students aged 9to 18 from six schools in Jiading District,Shanghai,for vision testing and questionnaire surveys from September to October of2021.The random forest model was utilized to rank the importance of myopia influencing factors,and the sliding window se-quential forward selection method was applied for dimensionality reduction.The factors with the highest importance scores and the lowest average out-of-bag estimation error rates were included in a multifactorial logistic regression analysis model to estimate the direction and effect values of the influencing factors.Results The myopia rate among the 1660 children and adolescents was 71.99%.The random forest model indicated that the average out-of-bag estimation error rate was minimized when the number of variables was six.When incorporated into the multifactorial logistic regression analysis model,the results showed that advancing educational stage(junior high:OR=5.191,95%CI:3.738-7.209;senior high:OR=6.989,95%CI:4.309-11.338),parental myopia(one parent myopic:OR=3.669,95%CI:2.723-4.943;both parents myopic:OR=5.231,95%CI:3.510-7.795),increased reading and writing duration(<1h:OR=12.192,95%CI:3.373-44.066;1-2h:OR=11.985,95%CI:3.364-42.704;≥2h:OR=10.547,95%CI:2.957-37.622),participation in extracurricular classes(<6 years:OR=3.114,95%CI:2.046-4.740;≥6 years:OR=1.817,95%CI:1.351-2.444),and using only ceiling lights for reading and writing at home after dark(OR=2.921,95%CI:2.009-4.248)were identified as risk factors for myopia.In contrast,a sleep duration of≥8 hours(OR=0.383,95%CI:0.264-0.557)was identified as a protective factor against myopia.Conclusion The prevalence of myopia among children and adolescents is high,and targeted measures should be implemented based on the identified in-fluencing factors for myopia prevention and control.

关 键 词：随机森林模型 儿童青少年 近视 影响因素 

分 类 号：R778.11[医药卫生—眼科]