Predicting stress and depressive symptoms using high-resolution smartphone data and sleep behavior in Danish adults

Study Objectives The early detection of mental disorders is crucial. Patterns of smartphone behavior have been suggested to predict mental disorders. The aim of this study was to develop and compare prediction models using a novel combination of smartphone and sleep behavior to predict early indicators of mental health problems, specifically high perceived stress and depressive symptoms. Methods The data material included two separate population samples nested within the SmartSleep Study. Prediction models were trained using information from 4522 Danish adults and tested in an independent test set comprising of 1885 adults. The prediction models utilized comprehensive information on subjective smartphone behavior, objective night-time smartphone behavior, and self-reported sleep behavior. Receiver operating characteristics area-under-the-curve (ROC AUC) values obtained in the test set were recorded as the performance metrics for each prediction model. Results Neither subjective nor objective smartphone behavior was found to add additional predictive information compared to basic sociodemographic factors when forecasting perceived stress or depressive symptoms. Instead, the best performance for predicting poor mental health was found in the sleep prediction model (AUC = 0.75, 95% CI: 0.72-0.78) for perceived stress and (AUC = 0.83, 95%CI: 0.80-0.85) for depressive symptoms, which included self-reported information on sleep quantity, sleep quality and the use of sleep medication. Conclusions Sleep behavior is an important predictor when forecasting mental health symptoms and it outperforms novel approaches using objective and subjective smartphone behavior.

Automated summary

It was found that sleep behavior is a significant predictor for forecasting mental health symptoms, outperforming novel approaches using objective and subjective smartphone behavior.