Roger Smith, an eighth-year student in Feinberg’s Medical Scientist Training Program, is developing an artificial intelligence tool to tie up medical loose ends identified in electronic health records. After graduating from the University of Notre Dame in 2014, he became a student in Feinberg’s Medical Scientist Training Program (MSTP).
Now rounding out his final year of medical school, Smith completed his PhD in the laboratory of Marc Mendillo, PhD, associate professor of Biochemistry and Molecular Genetics, where he investigated how cancer cells hijack processes necessary for protein homeostasis. He plans to pursue internal medicine research-track residency training with a fellowship in pulmonary and critical care medicine.
Read a Q&A with Smith below.
Why did you choose Feinberg?
Listen to Smith below:
I chose Northwestern’s Feinberg School of Medicine principally for the excellent clinical education. As an MD/PhD student I was also very interested in the robust, highly collaborative research environment that has been undergoing explosive growth in terms of NIH grants and private funding. Also, it’s located in the heart of Chicago, which provides access to world class music, food, arts and cultural events. And the icing on the cake for me was that it was close to family in the suburbs.
What inspired you to pursue an MD/PhD?
I was interested in med school through college, and I joined a research lab admittedly to support my medical school application. I really fell in love with the process of asking new questions and interpreting data to reveal some knowledge that is advancing the frontier of biology. Research impact can reach beyond the individual patient, potentially impacting tens of thousands of patients with new discoveries. MD/PhD training was an opportunity to integrate the wealth of knowledge through medical training with deliberate development of skills for asking tractable basic science research questions and devising rigorous experiments to test a hypothesis. PhD training also added practice in data interpretation, presentation and communication skills that complement the medical training. Combining these two skillsets allows me to ask better questions, especially when I see things in clinic environments that seem ripe for further investigation. It also gives me a better sense for what some practical limitations might be to translating basic science findings. Not to say that that should stifle innovation in any sense, but just having a better sense of how the pieces fit together across biomedicine.
What are your research interests?
I work in cancer biology, trying to understand how cancer cells co-opt processes for protein homeostasis to support their rapid proliferation.
The way I like to think about it is: if you’re a cancer cell rapidly dividing, you have to make all these proteins, and you need to make sure that they work well and go where they’re supposed to go along the way. It’s like being in a shipping department of Amazon around the holiday season. There’s explosive productivity and you need to make sure that there’s good quality control on all of the packaging. Protein homeostasis is an integral part of what allows cancer cells to be as aggressive as they are.
I’ve been broadly interested in how cells respond to stress. Whether that’s a cancer cell deprived of nutrients or oxygen, or how the lung is damaged and repaired during pneumonia. I have a growing interest in how the immune system toggles the process of inflammation to respond to an insult, whether that’s an infection or some sort of injury, and then how that process resolves when it’s time to not be inflamed anymore. Additionally, what the consequences of that process not working properly are on long-term health, particularly in the lung. Pursuing pulmonary and critical care fellowship training will allow me to dive into the clinical and research environments exploring these and other questions.
Tell us about your current research project.
It was fairly fortuitous that I found the informatics elective that introduced me to Dr. David Liebovitz. During a noon conference while on my medicine rotation, we had a discussion about opportunities for ChatGPT in the medical realm. As I was asking more questions, David Liebovitz recommended I take the informatics elective. So I took that class as an elective during my fourth year of medical school and it introduced me to the types of problems that already existed in clinical informatics and how the advent of user-friendly AI or large language models — tools like Chat GPT — can help address some of those.
I consider myself to be someone that’s oriented towards efficiency: I enjoy identifying and addressing inefficiencies that are both draining the users and dragging on a system. There are so many time-consuming tasks for medical providers in the electronic medical record. These are the same tasks that rank highly among the contributing factors for provider burnout.
Many of these, at least in theory, could be streamlined with technology, if not fully automated. I really see an opportunity there to both improve patient care and reduce the burden on providers, which I’m super excited about. I hope to see increasing investments in the development of these, certainly with patient safety and responsible use of data being a prerequisite. I think increasingly many people have talked to intelligent chat bots that can find information for them. Imagining what that might look like even in five or ten years, I hope that it can address a lot of these inefficiencies. My current research project is in the early tool-building stages.
A great way forward for these things and certainly a way to get started with them is to take existing tools and hook some sort of artificial intelligence/large language model up to it. In the electronic medical record at Northwestern, we have Epic, which is one of the largest providers. There is a tool that already exists in there called Slicer Dicer. This essentially allows you to look across the medical enterprise and identify patients with certain characteristics, or who have had certain diagnoses or certain lab tests that resulted in a certain way.
We have providers taking care of hundreds or thousands of patients throughout the year. Inevitably, for myriad reasons, through no one’s fault, things fall through the cracks. These are called “medical loose ends”. Whether it’s following up on an abnormal lab result or following through on a screening exam such as a mammogram or a colonoscopy, or things with potentially more grave consequences like therapeutic drug monitoring. For example, there are several medications including immunosuppressants or anti-seizure medications that require periodic drug monitoring to ensure proper therapeutic levels and minimize side effects. Currently, these things are left up to the provider to pay attention across thousands of patients.
Because all this knowledge is sitting there in the electronic medical record, we can use Epic’s Slicer Dicer tool to identify loose ends as we choose to define them. Once they’re identified, we have an ethical obligation to make sure they get addressed. There’s an opportunity to use these large language models and other AI tools to automate the identification and communication with patients on behalf of providers. We want to be really careful about adding any extra workload for them, because in many specialties they’re already near their limits, but you could imagine a scenario where the AI chat agent that’s being overseen by clinicians is able to contact patients and say, “Hey, you’re overdue for X, Y, Z. Please contact this office that is specifically designated for closing these medical loose ends – scheduling the appropriate labs or imaging studies and routing the results to the primary care physician.”
The project is in its early stages, but it’s exciting to think about how these tools are going to grow over time. I’m especially excited about this opportunity as a win-win, where we can give the care to patients that we’ve already said needed to be done and we can try to tie up those loose ends in an automated way.
What advice would you give to prospective medical students?
Figure out what excites you and pursue that. I’m in my eighth year of this program and it might sound like I have this nice clear narrative about how I did what I did and why I did it, but a lot of that is weaving a story back through after the fact. What got me into an MD/PhD to begin with was just following what was interesting and not worrying so much about some details that people can get caught up in.
Just be confident that you know yourself better than anybody else and pursue what seems the most important to you. If you’re doing that, you’ll have an edge because it matters to you more than other people. That includes picking a school to train; being close to family or friends or significant others or certain climates or outdoor activities. Those are all legitimate reasons to incorporate into a medical school decision.