Sleep disturbance recorded via wearable sensors predicts depression severity 9 years later

Comprehensive Summary

This study, conducted by Nur Hani Zainal and Peter F. Hitchcock, investigates sleep predictors in major depressive disorder (MDD) and their discovery via machine learning. Data utilized for this study was collected from 1054 participants, who were assessed through interviews and baseline physiological electrocardiograms at the beginning of the study and after 9 years. To investigate sleep markers, the researchers trained 8 machine learning models – least absolute shrinkage (LASSO), ridge, elastic net regularization (ENR), certification and regression trees (CART), random forest (RF), gradient boosting machine (GBM), support vector machine (SVM), and super learner (SLR) – which were then trained to predict major depressive disorder diagnosis at the 9-year follow-up. From the results, it was found that the model with the highest R2 value (0.198) and the lowest root mean squared error value (0.235) was GBM. Additionally, among all the measurements that were taken, it was found that the best 6 predictors of MDD diagnosis, from greatest to least, at the follow-up were: baseline MDD diagnosis, baseline generalized anxiety disorder diagnosis, baseline panic disorder severity, baseline total sleep time, baseline sleep efficiency, and baseline average sleep bouts during active phase. From the results, it was found that generalized anxiety disorder and panic disorder are correlated with major depressive disorder, and that certain sleep aspects - such as sleep time and efficiency - can be indicative of MDD risk.

Outcomes and Implications

Major depressive disorder (MDD) is a highly prevalent condition worldwide. It has been found to be correlated with autonomic nervous system conditions - such as high resting heart rate and low heart rate variability - and sleep features, which directly influence symptoms of major depressive disorder. With machine learning, these biomarkers of MDD can be utilized to assess disorder risk and help streamline prevention and treatment. Earlier detection of MDD using sleep features may lead to improvements in treatment outcomes