目的：筛选分析中小学生近视影响因素并建立预测模型，为儿童青少年的近视防控措施提供思路。

方法：于2023-09以学校为单位，采用分层整群抽样的方式抽取青岛市城区2所小学、2所初中、2所高中及1所职高中小学生1 759人，开展近视筛查和影响因素问卷调查。筛查和判定依据主要按照《标准对数视力表》(GB/T11533-2011)、《儿童青少年近视筛查规范》进行。基于机器学习算法LASSO联合XGBoost对近视影响因素分析及预测模型开发，以交互式Nomogram实现可视化。采用R统计软件版本4.3.3进行统计分析。

结果：青岛市城区中小学生生筛查性近视率为70.61%(1 242人)。筛选的最佳预测变量分别为年级、性别、父母是否近视、每天室内静坐时间、读写时眼睛距离书本超过一尺、每天的睡眠时间、看电视/玩电视游戏时眼睛距离电视显示屏的距离超过3 m、课间活动场所、补习班共多长时间、近距离用眼时多长时间休息一次眼睛、每天用电脑的时间、平均每天放学后做作业时间等12项影响因素。预测模型训练集、测试集的AUC分别为0.770(95%CI：0.751-0.789)、0.732(95%CI：0.714-0.750)。

结论：开发和验证了一个基于机器学习筛选近视影响因素的中小学生近视发病风险的预测模型，并实现了可视化。

METHODS:A total of 1 759 primary and secondary school students from 2 primary schools, 2 junior high schools, 2 senior high schools and 1 vocational high school in the urban area of Qingdao were sampled by means of stratified cluster sampling in September 2023. Vision screening and a questionnaire survey on influencing factors were carried out based on machine learning algorithms. The screening and determination were mainly conducted in accordance with the Standard Logarithmic Visual Acuity Chart(GB/T11533-2011)and the Specifications for Screening Myopia in Children and Adolescents. The influencing factors of myopia were analyzed and a prediction model was developed based on the machine learning algorithms LASSO in combination with XGBoost, and visualization was achieved through an interactive Nomogram. Statistical analysis was performed using R statistical software version 4.3.3.

RESULTS:The screening prevalence of myopia among primary and secondary school students in the urban area of Qingdao was 70.61%(1 242 cases). The optimal predictive variables for screening were grade, gender, whether parents were myopic, daily indoor sedentary time, appropriate distance between eyes and books during reading and writing, daily sleep time, distance between eyes and TV screen when watching TV/playing video games exceeding 3 meters, the playground during breaks, total duration of tutorial classes, how often eyes are rested during near work, daily computer usage time, and average daily homework time after school, totaling 12 influencing factors. The AUCs of the training set and test set were 0.770(95%CI:0.751-0.789)and 0.732(95%CI:0.714-0.750), respectively.

CONCLUSION: A machine learning-based prediction model was developed and validated to predict the risk of myopia onset in primary and secondary school students, accompanied by effective visualization techniques.