Prediction of depressive symptoms based on sleep quality, anxiety, and brain structure

Background: Depressive symptoms are rising in the general population and can lead to depression years later, but the contributing factors are less known. Although the link between sleep disturbances and depressive symptoms has been reported, the predictive role of sleep on depressive symptoms severity (DSS) and the impact of anxiety and brain structure on their interrelationship at the individual subject level remain poorly understood. Methods: Here, we used 1813 participants from three population-based datasets. We applied ensemble machine learning models to assess the predictive role of sleep, anxiety, and brain structure on DSS in the primary dataset (n = 1101), then we tested the generalizability of our findings in two independent datasets. In addition, we performed a mediation analysis to identify the effect of anxiety and brain structure on the link between sleep and DSS. Results: We observed that sleep quality could predict DSS (r = 0.43, rMSE = 2.73, R2 = 0.18), and adding anxiety strengthened its prediction (r = 0.67, rMSE = 2.25, R2 = 0.45). However, brain structure (alone or along with sleep/anxiety) did not predict DSS. Importantly, out-of-cohort validations of our findings in other samples provided similar findings. Further, anxiety scores (not brain structure) could mediate the link between sleep quality and DSS. Conclusion: Taken together, poor sleep quality and anxiety symptoms could predict DSS across three cohorts. We hope that our findings incentivize clinicians to consider the importance of screening and treating subjects with sleep and anxiety problems to reduce the burden of depressive symptoms.