UC San Diego

Mobile sensors are often used in health to track and monitor health, ranging from daily activities to diagnosing life-threatening conditions; however, they are underutilized for substance use and its disorders. Our work is focused on developing digital biomarkers from the physiological data captured from wearable devices for addiction. Specifically, we build models that combine the multimodal sensor data from wearable devices to detect drug administration, predict drug-induced mental states such as drug craving and euphoria. We further show that integrating drug pharmacokinetics into these data-driven models enhances the accuracy of drug monitoring, thereby increasing the generalizability and trust. A consistent pattern observed among these models was bias based on drug-usage history; therefore, we develop a model that screens users and distinguishes opioid misusers from prescription users, which would allow for more accurate prescription of opioids, minimizing the risk of addiction.