PASSPORT: PASsive Sensing to Prevent stress-related Overeating in Real-Time

PI: Nabil Alshurafa, PhD, associate professor of Preventive Medicine in the Division of Behaviorial Medicine

Summary: Stressful events, overeating and obesity have been associated with cardiovascular disease (CVD). Characterizing the relationship between these factors may inform targeted and timely interventions to prevent overeating episodes. To date most research relies on self-reports to identify stress and overeating episodes. Subjective self-reports are often retrospective and do not capture continuous physiological patterns to enable automated predictors of overeating. There is a need for objective approaches that continuously measure stress and overeating in real-time to further advance understanding of stress patterns that contribute to overeating. 

Wearable devices have become a powerful source for collecting health-related data using embedded sensors. Significant advances in technology have been made, including use of machine learning algorithms that process wearable data in real time. However, it is unclear how these real time indicators advance our understanding of the relationship between stress, overeating and obesity status. Moreover, most wearable devices collect sensor data in real time and process the data offline post hoc. We have developed a Band-Aid-like flexible wearable that can collect electrocardiography, photoplethysmography, and skin temperature data and an infrared (IR)-enabled camera that can collect both IR and color images.  

Building upon our preliminary research, the device can detect 1) stress using the Band-Aid-like device’s data and 2) eating using the IR- enabled camera’s data among a general population in free-living conditions. Data from these devices will then be used to build reliable and resilient (i.e., adaptive to the individual over time) machine-learned models that run in real time. Deploying machine learning algorithms with these novel features will likely improve stand- alone devices that can detect stress or overeating in real time, making them a viable option for timely intervention compared to existing wearable devices that simply collect sensor data for offline analysis.  

The aims of this study are to refine and deploy machine learning algorithms to detect stress (on a Band-Aid- like device) and overeating (on an IR-enabled camera) in real time. We will first assess the robustness of the stress detection model and refine the machine learning algorithms in a controlled setting where we can induce stress. Next, we will test the performance of these algorithms in real-world settings. We will then use this information to identify patterns of stress that can predict overeating. As an exploratory aim, we will determine if dynamically changing and personalizing the models to each individual improves model performance. 

This project has far reaching implications as it will further understanding of patterns in stress and overeating, and their relationship, to help predict overeating - thus providing fundamental knowledge about how we can deliver timely behavioral interventions to reduce CVD risk. 

Learn more about this project.