Detecting Mental Disorders with Wearables: A Large Cohort Study

Abstract

Depression and anxiety are among the most prevalent mental disorders, and they are usually interconnected. Although these mental disorders have drawn increasing attention due to their tremendous negative impacts on working ability and job performance, over 50% of patients are not recognized or adequately treated. Recent literature has shown the potential of using wearables for expediting the detection of mental health disorders, as physical activities are reported to be related to some mental health disorders. However, most prior studies on mental health with wearables were limited to small cohorts. The feasibility of detecting mental disorders in the community with a large and diverse population remains an open question. In this paper, we study the problem of detecting depression and anxiety disorders with commercial wearable activity trackers based on a public dataset including 8,996 participants and 1,247 diagnosed with mental disorders. The large cohort is highly diverse, spanning a wide spectrum of age, race, ethnicity, and education levels. While prior studies were usually limited to shallow machine learning models and feature engineering to accommodate the small sample sizes, we develop an end-to-end deep model combining a transformer encoder and convolutional neural network to directly learn from daily wearable features and detect mental disorders. WearNet achieves an area Under the Receiver Operating Characteristic curve (AUROC) of 0.717 (S.D. 0.009) and an AUPRC of 0.487 (S.D. 0.008) in detecting mental disorders while outperforming traditional and state-of-the-art machine learning models. This work demonstrates the feasibility and promise of using wearables to detect mental disorders in a large and diverse community.