The Human Algorithm

AI-powered research at UofL is transforming one-size-fits-all medicine

For most of us, artificial intelligence (AI) wasn’t a big part of our lives outside of science fiction until the debut of OpenAI’s ChatGPT tool in late 2022. And while the term “AI” is now firmly in our collective lexicon and being used in everyday tasks from developing easy weeknight dinner recipes to assistance in crafting the perfect grant proposal, many are still discovering the breadth of its benefits. But at the University of Louisville, researchers have used AI’s abilities to examine, explore and enhance health care for decades.

One such researcher is Adam Gaweda, associate professor in the Department of Medicine – Division of Nephrology and Hypertension, who has been utilizing AI to design clinical tools for real-world application at UofL since the early 2000s.

Now, he and fellow university researchers are making their approach to improve patient care personal.

“We’re continuing to find new ways AI systems can be implemented in specific domains,” Gaweda said. “Personalized treatment is a great use case for these techniques – when it’s done right.”

From standardized to specialized

Like all medical advancements, research typically begins by identifying a human problem and devoting time, energy, resources and innovative thinking toward seeking a solution. 

For Gaweda, that problem emerged in the clinic after years of observation revealed a central challenge in anemia treatment for late-stage kidney disease: a wide variation in individual patient responses.  Historically, patients had been given standard dosages, a biochemical version of one size fits all. Some patients were able to tolerate the standard dosage, but large numbers were not, putting them at risk of serious cardiovascular side effects such as stroke, thrombosis and heart attack.

“We knew it was vital for the physicians, the nephrologists, to have a way to better identify how each patient would respond individually to each drug,” Gaweda said. “So, we looked at several techniques of AI machine learning, including supervised learning, unsupervised learning and reinforcement learning. And through ten years of research, we came up with what we thought was optimal.”

 Gaweda and his team integrated elements from three AI learning techniques to build their model, then put it to the test in a randomized controlled trial at the UofL dialysis unit. They compared the AI-driven approach directly against standard dosage protocols used in patient care.

The results were everything the team and – most importantly – dialysis patients could have hoped for.

“We showed we could achieve optimal anemia management outcomes,” Gaweda said. “More importantly, we could do this at a much lower dosage level compared to the standard care.” 

The success of the program has opened the door for wider use of AI technology in the nephrology field, and today, the AI software developed at UofL is used by roughly 30% of the U.S. dialysis market.

Even with these successes, the push to innovate continues – and has expanded in scope.

The mind-body connection

Gaweda’s model sparked an exciting collaboration with another expert team at UofL who are joining forces to take the computational intelligence of what was learned in personalized kidney-disease management and applying it to the behavioral health domain.

Leading this charge is Cheri Levinson, a professor in the Department of Psychological and Brain Sciences and the Department of Pediatrics. From Levinson’s vantage point, UofL is on track to become a national center for AI-driven behavioral health innovation.

“Through cross-disciplinary collaboration – psychology, engineering, computer science, medicine – we are building tools that will allow clinicians everywhere to deliver precision care,” Levinson said. “What seemed unimaginable a decade ago is now within reach.”

Levinson believes AI is a pivotal tool in treating a range of mental health conditions, including the challenges she and her team encounter in UofL’s Eating Anxiety Treatment (EAT) Lab, where she serves as director. “Eating disorders fluctuate daily,” she said, “and AI gives us the ability to detect those changes and respond accordingly.”

This highlights how traditional approaches to care such as cognitive behavior therapy are being superseded by data-driven models that allow therapeutic measures to be as dynamic and personalized as the conditions themselves.

“Eating disorders are an ideal proving ground for AI-personalized mental health care,” Levinson said. “They are severe, heterogeneous and difficult to treat – exactly the type of illness where personalization is essential. To date, personalized behavioral treatment relied on sophisticated but limited idiographic models that could only be generated before treatment began, which means they could not update as symptoms changed. AI is changing that.”

Seeing people for who they are

Utilizing the computational models developed by Gaweda, Levinson’s team is developing algorithms with the ability to identify behavioral and emotional patterns in a patient that are impossible to detect clinically. This is done by analyzing thousands of data points and determining – with statistical precision – the symptoms that are maintaining the illness. From there, the team can recommend the best treatment module for the patient and adapt recommendations as symptoms shift.

But the most important factor in these AI-meets-behavioral-health initiatives is how they translate to the real world to improve both patient care and outcomes.

Irina Vanzhula, assistant professor in the Department of Psychological and Brain Sciences, underscored what is truly at stake in developing these therapies.

“A lot of people who come into our eating disorder trials tell us they have tried everything and nothing has worked for them,” Vanzhula said. “So, they are excited to try something different that covers the gaps that were missed before. Our vision for them is to build a personalized model of their symptoms that accounts for contextual factors, cultural factors and all the different conditions and diagnoses, and use modular treatment to match a set of interventions to that specific presentation.”

The life-changing potential of these and other personalized therapies created with the ethical use of artificial intelligence is not just something that can be shown in a model – it can be seen etched on the faces and heard in the voices of real-life UofL patients these researchers encounter every day.

“People feel they’re finally being seen and heard as individuals with unique challenges,” Vanzhula said. “And that is really the most powerful factor of all.”