Objective: To explore high-risk sexual behaviors of HIV/AIDS and related factors in young students in Guangzhou. Methods: A cross-sectional survey was conducted in 5 different types of Guangzhou colleges by convenience sampling with minimum number of classes per grade and 600 samples per school from September to November 2021. The R 4.2.2 software was used to consolidate databases. Simultaneously, a logistic regression model and a decision tree algorithm model, stratifying by whether sexual behaviors had occurred before, were constructed. In each layer, the prediction performance of the two models was evaluated through area under receiver operating characteristic and the confusion matrix, and then the model with high prediction performance was retained. Results: A total of 7 346 students were surveyed. The proportion of the respondents reporting sexual experience were 9.08% (667/7 346), in whom 26.24% (175/667) had risky sexual activity in the past year. The decision tree algorithm model performs well in predicting whether high-risk sexual behaviors have occurred in the past year. When the complexity parameter value is 0.018, and nsplit reaches 4, which means there are 5 leaf nodes in the model, the cross error of the tree will be the smallest. The first best grouping variable in the decision tree was whether to use condoms throughout the first sexual behavior. If condoms were used at their sexual debut, but homosexual practices have occurred in the past year, the probability of risky sexual behavior will increase. If homosexual practices have not occurred in the past year, but the age of sexual debut was below 18 years old while the period of HIV education was after high school, the probability of risk sexual behavior will also increase. Conclusions: AIDS-related risky behaviors of young students still deserved attention. The experience of sexual debut and whether AIDS-related health education has been received before the sexual debut were significant predictors for the occurrence of high-risk sexual behavior. The decision tree algorithm model has particular applicability for predicting and screening potential risk populations.
目的： 分析广州市青年学生艾滋病相关危险性行为的发生情况及影响因素。 方法： 2021年9-11月采用方便抽样方法在广州市5所不同类型大学开展横断面调查，每个年级至少抽取1个班级，每所学校样本量为600人。采用R 4.2.2软件进行数据整合、清洗和分析。以是否曾经发生过性行为为分层变量，同时构建logistic回归模型和决策树模型，每层通过受试者工作特征曲线下面积和混淆矩阵对两种模型预测性能进行评价，保留预测性能较高的模型。 结果： 共调查青年学生7 346人。自报曾经发生性行为者占9.08%（667/7 346），其中近1年发生危险性行为者占26.24%（175/667）。对于近1年是否发生危险性行为，决策树模型各项预测性能较高，当复杂度系数值为0.018，决策树生长层数为4，即有5个叶子结点时，树的交叉误差最小。决策树模型第一个最佳分组变量是首次性行为是否正确使用安全套，如果正确使用安全套但近1年发生男男性行为则发生危险性行为概率增加，如果近1年未发生男男性行为，但首次性行为年龄<18岁且开始接受艾滋病宣传教育的文化程度是高中及以上，发生危险性行为概率也会增加。 结论： 青年学生艾滋病相关危险性行为问题值得关注，首次性行为的经历、是否接受过艾滋病相关健康教育是后续发生危险性行为的重要预测因素。决策树模型对于预测和筛选HIV潜在风险人群具有一定适用性。.