湘西地区儿童青少年近视风险预测模型 构建

Construction of Myopia Risk Prediction Model for Children and Adolescents in Xiangxi Region

摘要: 目的：构建适用于湘西地区儿童青少年的近视风险预测模型，为区域性近视早期预警与二级预防提供实证依据。方法：采用横断面调查，纳入湘西地区943名9~12岁儿童青少年，构建涵盖人口学、家族史、生活行为、用眼习惯及环境六大维度的信息体系，采用多因素Logistic回归筛选近视风险因素并建立预测模型。结果：性别为女(OR = 2.185, 95%CI = 1.491~3.203)、体型肥胖(OR = 2.090, 95%CI = 1.199~3.640)、居住地为城镇(OR = 1.921, 95%CI = 1.298~2.844)、父母一方近视(OR = 2.400, 95%CI = 1.547~3.724)及父母双方近视(OR = 2.964, 95%CI = 1.533~5.731)是近视发生发展的危险因素；而课间休息场所为户外(OR = 0.531, 95%CI = 0.315~0.836)、晚上睡前不使用电子设备(OR = 0.592, 95%CI = 0.410~0.854)、不在暗光下看电子屏(OR = 0.616, 95%CI = 0.383~0.990)、读写距离约一尺(OR = 0.365, 95%CI = 0.232~0.574)及电视观看距离为屏幕对角线3~5倍(OR = 0.432, 95%CI = 0.269~0.693)是保护因素。模型测试集AUC为0.917，准确率为81.48%。结论：基于Logistic回归模型对儿童青少年近视风险具有良好的区分与预测能力，可支撑区域性近视早期筛查与精准干预。

Abstract: Objective: To establish a myopia risk prediction model suitable for children and adolescents in western Hunan, and to provide empirical evidence for regional early warning and secondary prevention of myopia. Methods: A cross-sectional survey was conducted, involving 943 children and adolescents aged 9 - 12 years in Xiangxi region. A health portrait information system covering six dimensions including demographics, family history, lifestyle, eye usage habits, and environment was established. Multivariate Logistic regression was used to screen myopia risk factors and build a prediction model. Results: Being female (OR = 2.185, 95%CI = 1.491~3.203), being obese (OR = 2.090, 95%CI = 1.199~3.640), living in urban areas (OR = 1.921, 95%CI = 1.298~2.844), having one parent with myopia (OR = 2.400, 95%CI = 1.547~3.724), and having both parents with myopia (OR = 2.964, 95%CI = 1.533~5.731) were identified as risk factors for myopia development; while having outdoor recess (OR = 0.531, 95%CI = 0.315~0.836), not using electronic devices before bedtime (OR = 0.592, 95%CI = 0.410~0.854), not viewing electronic screens in dim light (OR = 0.616, 95%CI = 0.383~0.990), maintaining a reading and writing distance of about one foot (OR = 0.365, 95%CI = 0.232~0.574), and keeping a TV viewing distance of 3~5 times the screen diagonal (OR = 0.432, 95%CI = 0.269~0.693) were protective factors. The AUC of the model test set was 0.917, with an accuracy rate of 81.48%. Conclusion: The Logistic regression model based on health portraits has a good ability to distinguish and predict myopia risk in children and adolescents, which can support regional early screening and precise intervention for myopia.