基于Lasso-Logistic回归构建高校大学生心理危机风险预测模型

Building a Predictive Model for Psychological Crisis Risk of College Students Based on Lasso-Logistic Regression

目的:探讨高校大学生心理危机风险的影响因素,构建并验证危机风险预测模型。方法:以2019年9月至2021年4月入组的浙江大学心理中心初始访谈的1879名大学生为研究对象,运用单因素Logisc回归和Lasso回归筛选危机风险的特征变量。采用随机抽样法以7∶3的比例将总样本分为训练集和验证集,并在训练集中基于筛选出的危机风险的特征变量构建多因素Logisc回归分析模型,并在训练集和验证集中分别验证该模型的区分度。结果:本研究具有危机风险的大学生255例(13.4%)。单因素Logistic回归和Lasso回归筛选出6个危机风险的相关变量,再经过多因素Logistic回归分析显示,“人际关系”(p=0.007)、“学院”(p<0.05)、“在校身份”(p≤0.001)、“自杀想法”(p<0.001)、“物质使用”(p=0.005)5个因素为危机风险的独立危险因素。验证分析结果提示该预测模型区分度良好,曲线下面积(area under the curve,AUC)为0.834。结论:本研究构建的预测模型具有较好的区分能力,可用于指导高校心理健康工作中对存在危机风险的高风险学生的早期识别和及时干预。

Objective:To explore the influencing factors of psychological crisis risk in college students and to construct and validate a crisis risk prediction model.Methods:Using 1879 college students from initial interviews enrolled in the Psychological Centre of Zhejiang University from September 2019 to April 2021 as research subjects,univariate logistic regression and Lasso regression were used to screen characteristic variables of crisis risk.A random sampling method was used to divide the subject into a training set and a validation set at a ratio of 7:3.A multi-factor logistic regression analysis model was constructed in the training set,and the discriminative ability of this model was verified separately in the training and validation sets.Results:In this study,there were 255(13.4%)cases of college students at risk of crisis.Univariate logistic regression and Lasso regression were used to screen out six relevant variables of crisis risk.The results of the multi-factorial regression analysis showed that‘interpersonal relationships’(p=0.007),‘college’(p<0.05),‘student status’(p≤0.001),‘suicidal thoughts’(p<0.001),and‘substance use’(p=0.005)were independent risk factors for crisis risk.The validation analysis results indicated that the predictive model had good discrimination,with an area under the curve(AUC)of 0.834.Conclusion:The risk prediction model constructed in this study has good discriminative ability and can be used to guide the early identification and timely intervention of high-risk students with crisis risk in mental health work in colleges.