Zhang Lirong, Zhao Shaocong, Yang Zhongbing, Zheng Hua, Lei Mingxing
Department of Physical Education, Xiamen University of Technology, Xiamen, Fujian, 361024, People's Republic of China.
School of Physical Education, Guizhou Normal University, Guizhou, 550025, People's Republic of China.
Psychol Res Behav Manag. 2024 Mar 13;17:1057-1071. doi: 10.2147/PRBM.S448698. eCollection 2024.
Sleep problems are prevalent among university students, yet there is a lack of effective models to assess the risk of sleep disturbance. Artificial intelligence (AI) provides an opportunity to develop a platform for evaluating the risk. This study aims to develop and validate an AI platform to stratify the risk of experiencing sleep disturbance for university students.
A total of 2243 university students were included, with 1882 students from five universities comprising the model derivation group and 361 students from two additional universities forming the external validation group. Six machine learning techniques, including extreme gradient boosting machine (eXGBM), decision tree (DT), k-nearest neighbor (KNN), random forest (RF), neural network (NN), and support vector machine (SVM), were employed to train models using the same set of features. The models' prediction performance was assessed based on discrimination and calibration, and feature importance was determined using Shapley Additive exPlanations (SHAP) analysis.
The prevalence of sleep disturbance was 44.69% in the model derivation group and 49.58% in the external validation group. Among the developed models, eXGBM exhibited superior performance, surpassing other models in metrics such as area under the curve (0.779, 95% CI: 0.728-0.830), accuracy (0.710), precision (0.737), F1 score (0.692), Brier score (0.193), and log loss (0.569). Calibration and decision curve analyses demonstrated favorable calibration ability and clinical net benefits, respectively. SHAP analysis identified five key features: stress score, severity of depression, vegetable consumption, age, and sedentary time. The AI platform was made available online at https://sleepdisturbancestudents-xakgzwectsw85cagdgkax9.streamlit.app/, enabling users to calculate individualized risk of sleep disturbance.
Sleep disturbance is prevalent among university students. This study presents an AI model capable of identifying students at high risk for sleep disturbance. The AI platform offers a valuable resource to guide interventions and improve sleep outcomes for university students.
睡眠问题在大学生中普遍存在，但缺乏有效的模型来评估睡眠障碍风险。人工智能（AI）为开发一个评估风险的平台提供了机会。本研究旨在开发并验证一个人工智能平台，用于对大学生经历睡眠障碍的风险进行分层。
共纳入2243名大学生，其中来自五所大学的1882名学生组成模型推导组，来自另外两所大学的361名学生组成外部验证组。采用六种机器学习技术，包括极端梯度提升机（eXGBM）、决策树（DT）、k近邻（KNN）、随机森林（RF）、神经网络（NN）和支持向量机（SVM），使用同一组特征训练模型。基于区分度和校准评估模型的预测性能，并使用夏普利值附加解释（SHAP）分析确定特征重要性。
模型推导组睡眠障碍的患病率为44.69%，外部验证组为49.58%。在开发的模型中，eXGBM表现出卓越的性能，在曲线下面积（0.779，95%CI：0.728 - 0.830）、准确率（0.710）、精确率（0.737）、F1分数（0.692）、布里尔分数（0.193）和对数损失（0.569）等指标上超过其他模型。校准分析和决策曲线分析分别显示出良好的校准能力和临床净效益。SHAP分析确定了五个关键特征：压力得分、抑郁严重程度、蔬菜摄入量、年龄和久坐时间。人工智能平台可在https://sleepdisturbancestudents-xakgzwectsw85cagdgkax9.streamlit.app/在线获取，用户可以计算个体睡眠障碍风险。
睡眠障碍在大学生中普遍存在。本研究提出了一个能够识别睡眠障碍高风险学生的人工智能模型。该人工智能平台为指导干预措施和改善大学生睡眠结果提供了宝贵资源。
