Scholarlink

Objective <p>Sleep disturbances are severe in adolescents with mood disorders, yet accurate and interpretable prediction tools for this high-risk population are lacking. This study aimed to develop and validate an explainable machine learning model to predict sleep disturbances risk and identify key predictors in a large clinical cohort.</p> Methods <p>This cross-sectional study included 1,505 adolescents with mood disorders, recruited from multiple hospital centers across China. The Pittsburgh Sleep Quality Index Scale served as the primary outcome measure for assessing sleep disturbances. A wide range of potential predictors was evaluated, spanning five domains: socio-demographic, psychological, behavioral, familial, and educational factors. We used the Boruta algorithm for feature selection, compared six machine learning models, and employed SHAP for model interpretation.</p> Results <p>The Random Forest model demonstrated superior performance, achieving an AUC of 0.941 (95% CI: 0.915–0.966), with an accuracy of 89.56%, a sensitivity of 0.980, and a specificity of 0.350. SHAP analysis identified psychological suffering and perceived discrimination in school as the most significant risk factors, while psychological resilience and core self-evaluations emerged as key protective factors. Additionally, the analysis indicated a non-linear impact of these factors on individual risk.</p> Conclusion <p>An explainable machine learning model exhibited high predictive performance in identifying the risk of sleep disturbances among adolescents with mood disorders. Our findings underscore the importance of both internal psychological states and adverse school experiences as critical intervention targets. This approach establishes a scientific foundation for clinical tools that facilitate the transition from generic advice to personalized, mechanism-targeted care, thereby advancing precision mental healthcare for youth.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An interpretable machine learning model for predicting sleep disturbances in adolescents with mood disorders: the key role of psychological factors

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月3日

查看全文
  • Xiaohong Chu,
  • Xudong Yang,
  • Zixin Ye,
  • Xinyu Hu,
  • Lili Chen,
  • Yuanping Deng,
  • Yawen Zheng,
  • Li Chen

摘要

Objective

Sleep disturbances are severe in adolescents with mood disorders, yet accurate and interpretable prediction tools for this high-risk population are lacking. This study aimed to develop and validate an explainable machine learning model to predict sleep disturbances risk and identify key predictors in a large clinical cohort.

Methods

This cross-sectional study included 1,505 adolescents with mood disorders, recruited from multiple hospital centers across China. The Pittsburgh Sleep Quality Index Scale served as the primary outcome measure for assessing sleep disturbances. A wide range of potential predictors was evaluated, spanning five domains: socio-demographic, psychological, behavioral, familial, and educational factors. We used the Boruta algorithm for feature selection, compared six machine learning models, and employed SHAP for model interpretation.

Results

The Random Forest model demonstrated superior performance, achieving an AUC of 0.941 (95% CI: 0.915–0.966), with an accuracy of 89.56%, a sensitivity of 0.980, and a specificity of 0.350. SHAP analysis identified psychological suffering and perceived discrimination in school as the most significant risk factors, while psychological resilience and core self-evaluations emerged as key protective factors. Additionally, the analysis indicated a non-linear impact of these factors on individual risk.

Conclusion

An explainable machine learning model exhibited high predictive performance in identifying the risk of sleep disturbances among adolescents with mood disorders. Our findings underscore the importance of both internal psychological states and adverse school experiences as critical intervention targets. This approach establishes a scientific foundation for clinical tools that facilitate the transition from generic advice to personalized, mechanism-targeted care, thereby advancing precision mental healthcare for youth.