潮汕地区农村留守中学生多重伤害预测模型构建

Establishment of a risk prediction model for poly-victimization among rural left-behind children

目的构建潮汕地区农村留守中学生多重伤害(PV)的风险预测模型,并评估模型的预测效果,为早期识别并预防学生PV的发生提供数据依据。方法于2020年1月—2021年9月,采用分层随机整群抽样方法在汕头市和揭阳市农村地区选取7所中学1005名留守学生进行问卷调查。调查内容包括个人、家庭、外部环境因素和心理因素(心理韧性、应对方式、自尊、领悟社会支持)及PV情况。运用R软件、采用Logistic回归筛选预测变量构建风险预测模型,使用ROC曲线下面积(AUC)、准确率、精确度、召回率、F1值和校准曲线对模型进行效果评价。结果留守中学生PV报告率为23.38%。Logistic回归分析结果显示,身体患病或残疾(β=1.02)、最近一年留级(β=1.31)、无亲密伙伴(β=1.00)、自伤意图(很少:β=0.58,偶尔:β=0.79)、同伴不良行为(β=0.90)、家庭成员吸烟(β=0.59)、父母/看护人违法犯罪史(β=1.04)、旁观过校园欺凌(β=0.78)、搬家(β=0.58)、采用发泄(β=0.34)和忍耐(β=0.28)的应对方式与潮汕地区留守中学生发生PV均呈正相关,心理韧性中的家庭支持(β=-0.31)与PV呈负相关(P值均<0.05)。对纳入多因素分析有意义的变量构建列线图预测模型,预测模型AUC为0.88,准确度为82.00%,精确度为77.78%,F1值为43.75%,校准图拟合良好,模型具有较好的区分度和校准度。结论构建的留守中学生PV的风险预测模型预测效能较好,有利于学校和社区早期识别发生PV的留守中学生。

Objective To construct a risk prediction model for poly-victimization(PV)among rural left-behind middle and high school students in Chaoshan,and to evaluate the prediction effect of the model,so as to provide scientific basis for early identification and prevention of PV among students.Methods A questionnaire survey was conducted among 1005 left-behind students,selected from 7 middle and high schools in rural areas of Shantou City and Jieyang City by a stratified random cluster sampling method from January 2020 to September 2021,for the personal,family,external environmental factors,psychological factors(mental resilience,coping approaches,self-esteem and social support)and PV situations.R software and Logistic regression were used to screen predictor variables to build a risk prediction model,and the area under the ROC curve(area under the curve,AUC),accuracy,precision,recall,F1 value and calibration curve were used to evaluate the model's effect.Results The incidence rate of PV among left-behind middle and high school students was 23.38%.The results of Logistic regression analysis showed that physical illness or disability(β=1.02),grade retention during the past year(β=1.31),having no close partner(β=1.00),self-harm intention(seldom:β=0.58,occasionally:β=0.79),negative peer behavior(β=0.90),family member smoking(β=0.59),criminal offenses of parents(β=1.04),witnessing school bullying(β=0.78),house moving(β=0.58),using venting(β=0.34)and the coping style of patience(β=0.28)were positively correlated with PV among left-behind children in Chaoshan area,and family support in psychological flexibility(β=-0.31)was negatively correlated with PV(P<0.05).A nomogram prediction model was constructed for the meaningful variables included in the multivariate analysis,and the prediction model AUC was 0.88,the accuracy was 82.00%,the precision was 77.78%,and the F1 value was 43.75%.The calibration plot fitted well,and the model had good discrimination and calibration.Conclusion The risk prediction model for left-behind middle and high school students with PV has good predictive performance and is helpful for schools and communities to early identify high-risk middle and high school students with PV.