Identifying Older Adults at Risk for Future Poor Sleep Quality: A Multidimensional Predictive Framework.

OBJECTIVE: To identify multidimensional risk factors associated with poor sleep quality and to develop and temporally validate a machine learning-based model for predicting the 2-year risk of poor sleep quality among community-dwelling older adults in China. METHODS: Data were drawn from the China Health and Retirement Longitudinal Study (2011-2018). Participants aged ≥ 60 years with good baseline sleep were included. Missing data were imputed using a random forest-based iterative method, and key predictors were selected using LASSO regression. Multiple machine learning algorithms were trained with cross-validation and evaluated using discrimination, calibration, and clinical utility metrics, with SHAP used for model interpretation. RESULTS: Among 3,471 participants, 23.2% developed poor sleep quality. Fifteen predictors across demographic, biological, psychological, and social-behavioral domains were identified. LightGBM performed slightly better overall (AUC = 0.641). CONCLUSIONS: The model demonstrated acceptable predictive performance and potential utility for early identification and targeted intervention. CLINICAL IMPLICATIONS: Poor sleep quality in older adults reflects the combined contributions of demographic, biological, psychological, and social factors, underscoring the need for multidimensional assessment framework in clinical and community settings. Identifying cumulative risk across emotional, physical, and social domains may facilitate earlier detection of high-risk individuals and more targeted prevention and intervention strategies.