Using Data Science to Predict Sleep Stages and Improve Patient Care

Last spring, Alice Chung ’21 (Neuroscience), Abhi Jadhav ’21 (Biomedical Engineering), Rohin Maganti ’21 (Biomedical Engineering and Global Health) and Kamyar Yazdani ’20 (Biology and Computer Science) received a Bass Connections Student Research Award to examine how a self-designed sleep-tracking device could be used to raise sleep awareness in centers of care. The wearable technology – Remedy – is designed to prevent adverse patient outcomes associated with sleep deprivation. Susanne Haga, Ziad Gellad, Sayan Mukherjee and Ann Saterbak serve as the group’s faculty mentors.

By Kamyar Yazdani, Rohin Maganti, Abhi Jhadav and Alice Chung

Since receiving the Bass Connections Student Research Award, we have conducted extensive research into different wearables’ ability to track sleep. Our research has shown that algorithms used to classify sleep stages in several wrist wearable manufacturers such as Fitbit and Apple are proprietary, and third-party applications such as ours are not able to access such data. As a result, we have shifted our focus to a data science project to use acceleration and heart rate data measured by such wearables to predict sleep stages (e.g., light, deep, awake, REM). As such, we added an additional mentor to our project. Dr. Sayan Mukherjee of Statistical Science has been guiding us through time-series models that can predict sleep stages most accurately.

This fall, we used the MESA (Multi-Ethnic Study of Atherosclerosis) dataset to conduct a survey of all potential models that could best give us a reliable prediction. We also validated our models on other datasets published online to ensure our model’s generalizability to different wearables and age groups.

Now, we are developing a Fitbit or Apple Watch application that uses our newly developed sleep stage classification algorithm to communicate with our Bluetooth hardware and traffic light for our first functional prototype. We are also planning to conduct clinical validation studies.

To demonstrate how our Remedy prototype works, our team also made a set of demo videos: Remedy Sleep Monitoring System Demo and Remedy Sleep Tracking System Methodology. These videos show how our algorithm uses heart rate, motion and light data from a wearable to predict patients’ sleep stage. A red, yellow or green light “traffic light” indicator then communicates the sleep stage to hospital staff and caretakers, who can use this information to avoid potential sleep disturbances.

Learn More

• Apply for a Bass Connections Student Research Award by March 6, and see what the other 2019 awardees are doing.
• Check out the 2020-20201 project teams, and apply by February 14 at 5:00 p.m.
• Explore Bass Connections summer programs.