摘要目的:探讨青少年心理健康状况对手机成瘾的影响，并采用极端梯度提升算法(XGBoost)和多因素Logistic回归模型，构建手机成瘾的预测模型。方法:2023年4月，对14 666名青少年进行横断面调查。采用自编的一般情况调查问卷、中学生心理健康问卷(MSSMHS)、青少年自我伤害行为问卷(ASHS)、交往焦虑量表(IAS)、手机成瘾指数量表(MPAI)、中学生羞耻感问卷(MSSS)、UCLA孤独量表(UCLA-LS)、同伴侵害量表(MPVS)、基本心理需求满足量表(BPNS)对所有参与者进行系统评估。使用R 4.3.2软件将参与者按7∶3的比例随机分为训练集和验证集，构建XGBoost模型和多因素Logistic回归模型，预测手机成瘾风险，并绘制列线图。通过Hosmer-Lemeshow检验、曲线下面积(AUC)和准确度(ACC)评价模型性能。结果:(1)最终纳入本研究的高中生人数为14 036人，其中5 069人(36.1%)存在手机成瘾；训练集中包括9 826名高中生，其中3 549人(36.1%)为手机成瘾者；验证集中4 210名高中生，其中1 520名(36.1%)为手机成瘾者。(2)XGBoost模型结果表明，羞耻感和社交焦虑是影响手机成瘾的两大主要预测因素。(3)多因素Logistic回归分析结果显示，焦虑[ B＝0.328， OR(95% CI)＝1.39(1.07～1.81)， P＝0.015]、人际关系敏感[ B＝0.311， OR(95% CI)＝1.36(1.05～1.77)， P＝0.018]、学习压力[ B＝0.606， OR(95% CI)＝1.83(1.46～2.31)， P<0.001]、情绪不稳定[ B＝0.775， OR(95% CI)＝2.17(1.70～2.78)， P<0.001]、社交焦虑[ B＝0.024， OR(95% CI)＝1.02(1.01～1.04)， P<0.001]、羞耻感[ B＝0.049， OR(95% CI)＝1.05(1.04～1.06)， P<0.001]、同伴侵害[ B＝0.037， OR(95% CI)＝1.04(1.02～1.06)， P<0.001]是手机成瘾的影响因素。(4)XGBoost模型在训练集中的ACC值和AUC值分别为0.890和0.929，在验证集中分别为0.865和0.864；多因素Logistic回归模型在训练集中的ACC值和AUC值分别为0.870和0.854，而在验证集中分别为0.867和0.859。 结论:焦虑、人际关系敏感、学习压力、情感不稳定、社交焦虑、羞耻感及同伴侵害等因素是高中阶段青少年手机成瘾的重要风险预测因素。

abstractsObjective:To explore the impact of adolescent mental health status on smartphone addiction, and construct a predictive model for smartphone addiction based on the eXtreme Gradient Boosting(XGBoost) algorithm and multivariate Logistic regression.Methods:In April 2023, a cross-sectional survey was conducted among 14 666 adolescents.All participants were systematically evaluated using a self-developed general information questionnaire, the middle school student mental health scale(MSSMHS), the adolescents self-harm scale(ASHS), the interaction anxiousness scale(IAS), the mobile phone addiction index(MPAI), the middle school students shame scale(MSSS), the UCLA loneliness scale(UCLA-LS), the multidimensional peer victimization scale(MPVS), and the basic psychological needs scale(BPNS).R software version 4.3.2 was used for data analysis. Participants were randomly divided into training set and validation set at the ratio of 7∶3.The XGBoost model and multivariate logistic regression model were constructed to predict the risk of smartphone addiction, and a nomogram was plotted.Model performance was evaluated using the Hosmer-Lemeshow test, area under the curve(AUC), and accuracy(ACC).Results:(1) A total of 14 036 high school students were included in the study, with 5 069(36.1%) exhibited smartphone addiction.The training set comprised 9 826 students, with 3 549(36.1%) being smartphone addicts.The validation set included 4 210 students, with 1 520(36.1%) being smartphone addicts.(2) The XGBoost model identified shame-proneness and social anxiety as the two main predictors of smartphone addiction.(3) Multivariate Logistic regression analysis revealed that anxiety( B=0.328, OR(95% CI)=1.39(1.07-1.81), P=0.015), interpersonal sensitivity( B=0.311, OR(95% CI)=1.36(1.05-1.77), P=0.018), learning pressure( B=0.606, OR(95% CI)=1.83(1.46-2.31), P<0.001), mood swings( B=0.775, OR(95% CI)=2.17(1.70-2.78), P<0.001), social anxiety( B=0.024, OR(95% CI)=1.02(1.01-1.04), P<0.001), shame-proneness( B=0.049, OR(95% CI)=1.05(1.04-1.06), P<0.001), and peer victimization( B=0.037, OR(95% CI)=1.04(1.02-1.06), P<0.001) were significant predictors of smartphone addiction.(4) The ACC and AUC values of the XGBoost model were 0.890 and 0.929 in the training set, and 0.865 and 0.864 in the validation set, respectively.The multivariate Logistic regression model achieved ACC and AUC values of 0.870 and 0.854 in the training set, and 0.867 and 0.859 in the validation set, respectively. Conclusion:Anxiety, interpersonal sensitivity, learning pressure, mood swings, social anxiety, shame-proneness, and peer victimization are identified risk predictors of smartphone addiction in high school adolescents.