This study investigated temporal changes in sleep quality among Chinese boarding high school students across four phases (2019:pre-pandemic baseline, 2020:outbreak containment, 2021:dynamic mitigation, and 2024:post-pandemic adaptation), aiming to identify optimal PSQI cutoffs and key predictors using machine learning. A total of 4,592 Grade 11 students from Hebei Province were surveyed using a four-wave repeated cross-sectional design. Sleep quality was assessed using the Pittsburgh Sleep Quality Index (PSQI). Generalized additive models and XGBoost were used to evaluate trends and predictors. SHAP analysis quantified feature contributions. PSQI scores peaked in 2019 (M = 5.64) and improved significantly in 2020 (Δ = - 0.31, P = 0.012), with continued improvement through 2024. Machine learning identified PSQI = 5 as the optimal threshold (AUC = 0.755; accuracy = 0.699). BMI-age interaction was the most influential predictor, followed by height, age, and sex. Findings support updates to adolescent sleep guidelines and demonstrate the utility of machine learning for risk stratification in future public health contexts.
Zhang et al. (Thu,) studied this question.