Research & Innovation

Stevens Student Seeks the Sweet Spot in Measuring the Effect of Stress on Glucose Levels

Michelle Morrone’s Senior Design machine-learning project could help people with type 1 diabetes use electrodermal activity to factor stress into their glucose monitoring for better disease management

Glucose monitoring kit

For people with type 1 diabetes, regulating blood glucose levels is a constant, complex, and often nerve-wracking balancing act of monitoring data such as daily insulin shots, food intake, and physical activity, as well as the individual’s harder-to-quantify emotional state. In her Senior Design project, Stevens Institute of Technology chemical biology major Michelle Morrone ’21 aims to help take some of the stress out of the process by using the body’s electrodermal activity (EDA) and machine learning models to account for stress in glucose forecasting and help better predict how much insulin to administer through insulin pumps.

ENTER ALT TEXT
Michelle Moronne '21

“Stress is complicated, but it can do just as much to glucose levels as eating a meal or exercising,” Morrone said. “That makes it difficult to accurately and quickly calculate and predict a person’s glucose levels to administer the right amount of insulin. The goal of my project is to use machine learning to find the link between stress, which is reflected in the skin’s EDA, and glucose levels. Eventually, this data could be integrated with automated insulin pumps and really improve the lives of people with Type 1 diabetes.”

Understanding the biology and biometry of stress, diabetes, and glucose levels

While some studies have examined the relationship of glucose and stress, using machine learning and EDA levels, obtained by body sensors, adds a new twist to the equation.

“Michelle took what we were doing in the Health and AI Lab around stress and diabetes, and stress and glucose, and put her own spin on it,” said her advisor, Samantha Kleinberg, associate professor of computer science. “It’s a great example of how Stevens encourages and supports interdisciplinary research, and research at the undergraduate level.”

With Morrone’s academic concentration in bioinformatics and her minor studies in computer science, it seems the perfect research challenge.

“As a college student, I am stressed a lot of the time, and coming from the biology side, I know that can do crazy things to your body and affect your health,” she explained. “One of those things is raising glucose levels for the fight-or-flight instinct. I thought that it probably makes it more challenging for patients with diabetes to manage. If you're stressed out about midterms, realizing you're stressed out and then realizing that you need to adjust your insulin for that just adds to the burden and the amount of time you're spending managing your diabetes. It would be great if that could be automatically done for you.”

Morrone began her research last fall with literature reviews related to stress, diabetes, and glucose levels to understand the biology involved, while also reading up on machine learning for predicting glucose levels. This semester, she’s been exploring machine learning algorithms to account for stress and glucose levels based on the OhioT1DM dataset, which contains blood glucose, insulin, life-event, and physiological fitness band data from people with type 1 diabetes on insulin pump therapy with continuous glucose monitoring.

“I'm using machine learning algorithms to predict glucose levels, and using parameters like stress and heart rate, in addition to other parameters that have been used before such as insulin levels and meals, to see if that improves the accuracy of the prediction,” she said. “By graphing the data, I’m looking for patterns like glucose spikes and EDA spikes that occur around the same time with no other explanation, to see if there's any correlation.”

Making machine learning work for real-life education and real-life solutions

Although Morrone’s no novice to programming, machine learning has been a whole new world for her.

“I'm still learning how to code machine learning algorithms,” she said. “I’ve also had to learn how to convert the data from its original XML format into a CSV format that I'm able to manipulate and transform into visual graphs, which is also something I haven't done before. It can be frustrating when I'm not quite sure what I'm doing, but I've been working with other students working on diabetes research in the lab, and it helps to realize they also have some of the same questions as I do. We put our heads together to figure out what things mean, figure out the documentation, and continue looking at the programming to help figure it out. It's been helpful knowing that I'm not alone. Even though we've all come from different experiences and they're all grad students, we're coming together to help each other figure it out.”

In fact, since undergrad Morrone has become the resident expert on the OhioT1DM dataset, she has been able to help the graduate students work with the data as well.

“Even though we each have our own personalized spin on it, it’s been great that we can work on it together,” she said.

Collaboration is the legacy she hopes to leave for other Stevens students.

“From what I've learned so far, stress can have very little effect or a fairly decent effect, depending on whether it’s more long-term or acute, and positive or negative,” she noted. “Each one does something different to the body, but you can't tell the types of stress from the EDA, and the effects of stress on glucose levels are probably going to be different from person to person. I hope that my findings might benefit other students in our lab who are focused more on personalization, or who might be able to consider stress to account for complications in the data, so they can continue analyzing stress data to add it to the overall algorithm to help make accurate predictions of how much insulin needs to be given.”

She has also teamed up with other students in her Alpha Phi Omega service fraternity, where her Senior Design project helped inspire the idea for a diabetes innovation panel for National Service Week.

“When we started discussing what events we could potentially do, they knew that I was working with diabetes innovations, and that led to the idea of having a panel speak about what’s going on with diabetes studies,” she said. “It was interesting hearing about the different sides of diabetes research, and knowing that it could help the broader community.”

Morrone’s research also provides a solid foundation for her post-Stevens plans.

“I want to work with chemistry and biology, and I also want to work with computer science because I like seeing the intersection among these disciplines,” she said. “This project has taught me a lot about computational studies and computer programming and machine learning. I’m going to look for a job in this area, and I may go to grad school. Whatever I do, this project has been worthwhile for both teaching me things that I can hopefully apply to my future, and hopefully leading to devices that will improve the lives of people with diabetes in the future as well.”

ENTER ALT TEXT
Michelle Morrone (top right) explained a graph depicting continuous glucose monitor, EDA, and heart rate data to fellow diabetes researchers (from top to bottom) Louis Gomez, a computer science PhD candidate and member of the Class of 2024; Ayesha Parveen, master in computer science student and a member of the Class of 2022; and Prajwal Prakash, master in computer science student, ’21. CREDIT: Michelle Morrone

Learn more about chemistry and chemical biology at Stevens: