‘Push our boundaries’

The intersection of artificial intelligence and medicine — and the urgency to integrate it into medical education — was the primary focus of a special AI-themed day held during OUWB’s Medical Education Week.

The 13^th Annual William Davidson Medical Education Week was held May 13-17 and featured the theme of “Reimagining Medical Education.”

The week was jam-packed with special guests and presentations with one full day dedicated entirely to artificial intelligence (AI) and medicine.

Guest speakers were Erkin Ötleş, M.D., Ph.D., emergency medicine resident, University of Wisconsin; Cornelius James, M.D., assistant professor, Departments of Internal Medicine, Pediatrics, and Learning Health Sciences, University of Michigan Medical School; Gustavo Patino, M.D., associate dean, Undergraduate Medical Education, Western Michigan University Homer Stryker M.D. School of Medicine; and Ramin Homayouni, Ph.D., professor, Department of Foundational Medical Studies, OUWB.

Christopher Carpenter, M.D., Stephan Sharf Dean, OUWB, attended the event and said it’s a topic that the school simply can’t ignore. In fact, the idea for an AI day during medical education week originated with him.

“As in nearly every other segment of our society, the adoption of AI in medicine is accelerating, and we have to recognize its immense promise in helping improve health but also guard against its inappropriate application that could end up in extreme cases harming individuals,” he said.   

“I hope that everyone walked away with an appreciation for our need to continue to push our boundaries as educators, making informed efforts to enhance all aspects of medical education,” he added.

Introduction to Artificial Intelligence for Medicine – Lessons from Health Care

Ötleş presented the day’s first session that naturally centered on an introduction to the topic of AI in health care.

The emergency medicine resident previously was in the University of Michigan Medical School’s Medical Scientist Training Program (MSTP).  It’s a National Institute of Health-funded program designed to train physician-scientists by integrating medical education with scientific training.

Ötleş’s dissertation was on “Machine Learning for Healthcare: Model Development and Implementation in Longitudinal Settings.”

He said that he hoped attendees of his presentation at OUWB would better understand that physicians need to be actively engaged in shaping the continued development of AI, which he defined as “intelligence (perceiving, synthesizing, and inferring information) demonstrated by machines.”

“AI is not magic, it’s an engineered system…that means we have control over how it’s made, how it’s implemented, and how we use it in clinical practice,” he said. “As physicians, we should feel comfortable guiding all of those parts and be actively engaged.”

He explained that most people use AI in some way regularly, whether trying to find the best driving route to a destination or using “assistants” like Siri or Alexa to search the internet. Anyone who has ever used ChatGPT has used “generative AI” — AI that generates data (text, images, etc.) using generative models in response to prompts.

AI already is being used in medicine, too.

Ötleş defined AI in health care, provided better understanding of the connections between generative and predictive AI, showed examples of AI in clinical settings, and looked at potential future uses of the technology.

And it continues to gain momentum.

By example, he pointed to the growth in FDA-cleared AI devices for medical use. In 1995, there were only a handful. Last year, there were nearly 700.

“This is just the tip of the iceberg, too,” he said. “Because many devices that use AI don’t need to be certified by the FDA because they’re often being used for informational purposes and not actually doing diagnostics.”

But, he added, AI is not a typical part of medical education, and learners are “unprepared to use, assess, and develop AI tools.”

“We need to think about how we can address this,” he said.

That’s where the next session presented by James comes into play.

AI: The Next Paradigm Shift in Medical Education

James said he wanted everyone to take at least one thing away from his presentation: AI isn’t five or 10 years down the road — it’s here now.

“And the longer we want to begin to prepare our learners…for these technologies, the further behind we get, which means that we’re potentially not providing the best or most optimal care that we can provide for our patients,” he said.

James’ summarized the current state of AI in medical education, provided a vision for AI in medical education, talked about initiatives designed to begin preparing clinicians for AI in health care, and led a general discussion on related topics.

He used the metaphor of three speeding trains to represent AI in medical education. One represented the use of AI in teaching and learning purposes, another teaches learners to engage with those technologies, and a third uses AI for med ed scholarship.

“I 200% agree with Erkin on the importance of us needing to prepare clinicians to not only engage with these technologies, but to lead when it comes to the use of these technologies,” he said.

He talked about the kind of skills physicians of the future will need to provide data-driven, patient-centered, evidence-based care — and made it clear that medical education needs to embrace AI.

Currently, AI in medical education is primarily derived from electives, online courses/modules, workshops, certificate programs, or interest groups.  

Over the next one to two years, James said he expects to see AI integrated into courses where there is clear overlapping content. Such courses will include health systems sciences, clinical reasoning, evidence-based medicine, and clinical skills.

During the next five to 10 years, he said, data sciences will join clinical, basic, and health system sciences as the pillars of medical education.

“Sometimes when I present this, people ask what will have to be taken out of the curriculum,” he said. “Those are difficult conversations because I believe some things will have to be taken out.”

James hopes that the result will be that medical education will focus more on empathy, critical thinking, communication, teamwork, and innovation and less on information recall and administration.

Tying it all together

The rest of the day tied everything together with discussions and presentations that further addressed the promise and problems of AI.

Patino talked about “Interpretability and Biases within the AI Model.”  Specifically, he addressed the sources of bias in AI models and reviewed the role of interpretability techniques in controlling for bias and enhancing applicability of AI models. He also encouraged systematic efforts to understand the diverse aspects of AI models beyond overall performance. 

In short, he expressed that the primary issue stems from the fact that oftentimes AI relies on existing information that might be flawed and as a result, perpetuates any inherent issues with the original sources from which it draws.

The day wrapped with Homayouni’s presentation on “Machine Learning Using Clinical Notes: Applications in Population Health.”

Like the other presenters, Homayouni talked about the potential pitfalls of not embracing the role of AI in health care — and how AI can’t exactly replicate the kind of empathy and compassion that humans can express.

Homayouni also talked about the potential different uses for AI in population health: risk scoring for chronic disease; avoiding 30-day hospital readmissions; getting ahead of patient deterioration; and predicting patient utilization patterns, just to name a few.

One specific example addressed how AI can be used to identify social determinants of health by using information already in patient records.

“If we can predict who might need services…maybe we can lower the overall health system costs,” he said.

Overall, Carpenter said he was pleased with the outcome of the AI day.

“There was great participation and I believe a lot was gained that we can apply to educating our learners,” he said.