The Daily Tar Heel
Printing news. Raising hell. Since 1893.
Sunday, Feb. 25, 2024 Newsletters Latest print issue

We keep you informed.

Help us keep going. Donate Today.
The Daily Tar Heel

Q&A with UNC Professor who is working to use artificial intelligence in public health


In the past decade, there has been a push to integrate artificial intelligence into the health care setting. Michael R. Kosorok, a professor in the departments of Biostatistics and Statistics and Operations Research at UNC, discussed current efforts to use AI to improve public health and precision medicine. Kosorok has focused on using AI to improve decision making for individuals with health concerns. 

The Daily Tar Heel:  When did you first become interested in artificial intelligence and its relationship to healthcare?

Michael R. Kosorok:  Well, I’ve always been interested in AI since I first read about it in science fiction and other things as a younger person. However, I started shifting my own career, so I was focusing on AI and precision medicine especially about ten years ago, shortly after moving here from Wisconsin. 

DTH:  What does the term ‘AI’ mean to you? What are some examples of how it is currently used in health care?

MRK:  Right, so artificial intelligence is a class of methods of analyzing data that allow you to be able to learn something from the data; to be able to use it for prediction or for making decisions or for classification. You can use it to translate languages. You can use it to identify objects in a picture. There are many ways you can use it. You can construct logic tables so that you can make decisions using AI. 

For me, the most interesting aspects of AI are how AI can take data and convert that data into something that we can use for prediction or decision-making. In my work, I’ve largely used artificial intelligence to figure out, based on data that we collect either in real time or in batch mode, and be able to determine what is action or treatment that a patient receive that will result in the best outcomes for that patient. That’s different from prediction, because in prediction I would just be trying to figure out what happens to a patient. Rather than knowing what will happen, I want to know what I can do that will lead to a better result in the future.

A couple of years ago, AI was used to be able to come up with a method that could predict, fairly accurately, whether someone had skin cancer based on photos. It turns out that it appeared to perform as well as a dermatologist would, and that’s much, much (sic) more complex than an AI using logic tables. 

In my main research area — precision medicine — I’m trying to figure out how to make decisions for a person, patient, treatment and doctor. How they should act or what they should do, what actions they could take that will lead to a better clinical outcome at the next moment in time, or the next day, or next year or the long-term health result. That’s a very difficult challenge and a lot of interesting work has been going on in this area during the last five to 10 years. 

DTH:  Can you define precision medicine as it relates to AI and the applications of your research?

MRK: Precision medicine is really trying to determine how to treat, or act or how to apply anything related to care that involves decision-making. How do you do that in a way that leads to the best outcome? Historically, precision medicine has focused on trying to find a very narrow scope of individuals with a disease and a very specific mechanism of action for which you can develop a targeted drug to fix that problem. Our goal is to look at the whole population, divide that whole population into groups, within which each groups would get the same treatment. We don’t know in advance what those groups look like, and AI helps us do that. Our goal is to benefit everyone, not just a small subgroup. 

DTH:  Would you care to expand on some of the research you’re currently working on?

MRK:  I already mentioned the project that we’re working on with Type-1 diabetes. We are working on creating a type of AI-based decision support so a patient can manage their diabetes and also be able to successfully exercise. We’re not doing prediction here; we’re trying to figure out what actions people can take that will lead to better outcomes and to recommend those actions in a timely way. One of the key tools we’re using for the artificial intelligence is something called V-learning, which was developed here by several individuals at this institution and also at N.C. State. The core of the research was done at UNC by myself and Elizabeth Mayer-Davis in nutrition, and Eric Laber in the statistics department at N.C. State. 

Our goal is to have a working prototype by next summer.  

DTH: Are there people who reject the integration of AI and precision medicine in healthcare?

MRK: There are people who reject the concept of precision medicine because they think it focuses on narrow subgroups, but I’ve just refuted that by saying that’s not our focus, that we’re focused on the whole population. 

There are people that are uncomfortable with AI because it’s very mysterious, but my answer to that is as we study and work with that, we can learn how it works. A lot of my research is to prove, either mathematically or through simulations and applications, that AI can work and then show that it does. There are people in various fields who worry about it, and many of their questions are actually very helpful in the research process. Still, there are many people who reject it out of hand and it’s very difficult to work with them because in many cases there isn’t an alternative or better solution than AI. In many cases, AI really is the best solution to be had. 

As with all new things, there does need to be good scrutiny and peer review of these ideas. They do need to be evaluated carefully, so we work on studying things like: Are these methods reproducible? Will they work in a variety of environments? Reproducible research is very important to us, so we tend to scrutinize and evaluate the AI very carefully before we begin using it for human research. 

DTH:  Do you have any final thoughts you would like to add?

MRK: (AI and precision medicine) are not just permutations of things we’ve done before. There really is a confluence of new ideas that are very revolutionary, I think.

This interview has been edited for clarity and brevity. 


To get the day's news and headlines in your inbox each morning, sign up for our email newsletters.

Special Print Edition
The Daily Tar Heel Victory Paper for February 5, 2024