Featured Researcher: Monica Agrawal 

Center member Monica Agrawal, PhD, is an Assistant Professor of Biostatistics & Bioinformatics, Computer Science, and Biomedical Engineering at Duke. She joined Duke after she earned her PhD in computer science from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). 

Monica is clearly a rising star in the field of computational digital health. Her work bridges computer science, biostatistics, and medicine in ways that are both technically rigorous and deeply translational. We’re thrilled to have her at Duke and as part of the Center—she’s already making a strong impact on how we think about AI, healthcare, and real-world data.

Here’s what she’s thinking about and working on.

What excites you most about computational digital health innovation right now?

Computational and digital health are poised to reimagine the status quo of medicine and ultimately improve outcomes for patients. Real-time monitoring and increased care accessibility can usher in earlier diagnosis and more personalized evidence-based decision making.

Now in particular, we’re at an inflection point to see rapid innovation in this space. First, we have rich sources of health data: from wearables and omics to notes, labs, and imaging in electronic health record. These multimodal data enable researchers to piece together a more cohesive picture of patients’ journeys. Second, new technologies, like AI and large language models, enable us to scalably analyze and draw insights from this huge amount of data we’re collecting. Computational digital health could truly transform workflows for patients, clinicians, and researchers.

How did you get interested in this computational and digital health? 

I was always interested in biomedical research, but after working in a neuroscience lab in high school, I realized I was not meant to be a bench biologist. I then discovered I loved computer science during my freshman year of college, and I found myself seeking opportunities where I could learn more about the intersection of computer science and biomedical research.  My freshman summer, I worked in a biomedical optics lab which was my first real exposure; projects in the research group involved engineering mobile health diagnostics and designing medical computer vision algorithms. Then my junior year, once I had more computer science experience myself, I started researching computational methods for discovery of new disease-related proteins, and I interned at health technology company where I first developed natural language processing methods for electronic health record data. After those experiences, I found that I really loved working at this intersection, and that spurred me to apply for PhDs with computational health as the focus. 

What research are you currently involved in? 

At a high level, I develop natural language processing methods to understand biomedical text (e.g., clinical notes, medical literature) and use that understanding to improve workflows across healthcare. The data in electronic health records (EHRs) contain a gold mine of real-world evidence that could enable us to address questions in precision medicine, but much of the information we care about is trapped within unstructured free-text clinical notes. Some of my major focuses have been designing algorithms and human-AI systems to enable scalable understanding of this data and building smarter electronic health records. 

For the past four years, my research has revolved around large language models in medicine. More recently, this includes analyzing where online large language models learn clinical information — and clinical misinformation — from, and thinking through the challenges that come with monitoring and evaluating generative AI tools in healthcare. There’s a current disconnect between how we evaluate algorithms and how they get used. I’m particularly interested in studying how patients may already be using large language models to parse health information and receive health advice. I think it’s important to think through how we might need to change algorithms and user interfaces to account for unintended consequences. 

Finally, I have a lot of interest in questions in women’s health; as it has been traditionally understudied, AI is particularly poised to make a huge difference. 

How do you see AI impacting or advancing your work? 

As a computer scientist who works in natural language processing, AI is the bread and butter of everything I do. All of my work in computational health essentially falls under the buckets of understanding the mechanisms and pitfalls of current AI paradigms, subsequently developing new AI algorithms, and investigating human-AI interaction.  

What’s your outlook for the future? Where do you see the field being in 10 years? 

I’m really excited to see how paradigms change about how we acquire clinical knowledge, as we expand much further beyond what we can learn from traditional clinical trials alone.  There’s so much heterogeneity in how patients present with conditions and respond to treatments that we haven’t scratched the surface of. My hope is in 10 years we’re going to be seeing a lot more personalization in health guidelines and treatment recommendations, catalyzed by data-driven hypothesis generation. Additionally, I think there will be better monitoring to help coordinate across care providers, leading to earlier diagnosis and intervention when needed. However, this vision comes with a lot of complex open modeling challenges, e.g. the fusion of irregularly sampled multimodal data, that will require significant advancements in AI.  

What’s something people don’t know about you?

I love both hiking and chocolate, and once I hiked to see the mountain that is on the Toblerone bar. I did in fact eat some Toblerone while on that hike.

And I really love donuts. I bought a donut wall for my dissertation defense. If people have recommendations for the best donuts in the Triangle, I’ll totally take those.