Towards Device Agnostic Detection of Stress and Craving in Patients with Substance Use Disorder

Abstract

Novel technologies have great potential to improve the treatment of individuals with substance use disorder (SUD) and to reduce the current high rate of relapse (i.e. return to drug use). Wearable sensor-based systems that continuously measure physiology can provide information about behavior and opportunities for real-time interventions. We have previously developed an mHealth system which includes a wearable sensor, a mobile phone app, and a cloud-based server with embedded machine learning algorithms which detect stress and craving. The system functions as a just-in-time intervention tool to help patients de-escalate and as a tool for clinicians to tailor treatment based on stress and craving patterns observed. However, in our pilot work we found that to deploy the system to diverse socioeconomic populations and to increase usability, the system must be able to work efficiently with cost-effective and popular commercial wearable devices. To make the system device agnostic, methods to transform the data from a commercially available wearable for use in algorithms developed from research grade wearable sensor are proposed. The accuracy of these transformations in detecting stress and craving in individuals with SUD is further explored.