Associations of Physical Activity and Sleep Quality with Depression Risk in U.S. Adults: An Interpretable Machine-Learning Analysis

Background The increasing burden of depression highlights the importance of modifiable lifestyle factors, particularly physical activity and sleep. However, evidence regarding their independent dose–response relationships and thresholds remains limited. Methods A total of 21,201 participants aged 12 years or older from NHANES 2007–2014 were included. Total physical activity was quantified as the Total Metabolic Equivalent of Task (TMET, in MET-h/week), calculated as the sum of contributions from five distinct activity types. Sleep quality scores (ranging from 0 to 3) were constructed from sleep duration, self-reported sleep problems, and diagnosed sleep disorders. Depressive symptoms were defined as a Patient Health Questionnaire-9 (PHQ-9) score of 5 or above. Weighted logistic regression, restricted cubic splines, and two-piecewise linear regression models were used to assess dose–response relationships and thresholds. Subgroup analyses evaluated effect modifications. Variables selected by LASSO and Boruta were used to develop 12 machine learning models, with SHAP applied for model interpretation. Results Each 10 MET-h/week increase in TMET was linked to a 2% reduction in depression risk (OR = 0.98, 95% CI: 0.97–0.99). Sleep quality showed a linear inverse association, with the highest sleep score group having only 11% of the risk compared to the lowest group (OR = 0.11, 95% CI: 0.08–0.13). TMET exhibited an L-shaped relationship with depression risk, featuring a threshold at approximately 12 MET-h/week; below this threshold, each additional 10 MET-h/week was associated with a 34% reduction in depression risk (OR = 0.66, 95% CI: 0.59–0.72). The sleep–depression association was stronger in individuals aged 60 years or younger and in those without diabetes, while the protective effect of physical activity remained consistent across subgroups. The AdaBoost model demonstrated the best predictive performance (AUC = 0.78), and SHAP analysis revealed that sleep score was the most important predictor, followed by poverty income ratio, smoking, education, and TMET in descending order of importance. Conclusion Moderate-to-vigorous physical activity (especially below 12 MET-h/week) and high-quality sleep are independent protective factors against depressive symptoms, with sleep showing a stronger effect. Interventions should target low-activity populations by promoting regular exercise and improving sleep hygiene. Future longitudinal and wearable-based studies are needed to clarify causal mechanisms and support tailored prevention strategies.