The gift of life involves the recovery of organs for transplantation. But it is not always that simple. When a deceased organ donor is identified, a list is generated of transplant candidates who are suitable to receive each organ. Organ transplants require a compatible match and those matches can be rejected based on a poor fit between a donor and recipient. What many don’t realize is how complex this process is, especially how much work it takes and how critical timing is when making a compatible match between donor and recipient – ranging from 36 hours to only four, depending on how long the organ can remain preserved outside the body.
When predicting cross matches, medical professionals use a variety of tools for analyses. But what if there was a more refined way to better match donors and recipients, a technology with a simple user interface that leads to even faster accurate results?
This is what UNC-Chapel Hill researchers are pursuing – a new technology that leans on artificial intelligence plus their own expertise to make the best possible matches for patients in need.
Solving a complex answer in a short amount of time
Typically, when a solid organ recipient is matched for a deceased donor offer, the histocompatibility lab at the UNC Medical Center will review the aspects of the donor and the recipient to compare them digitally. From that comparison, lab director Eric Weimer would make a recommendation to the transplant team of how compatible the pair is for transplant.
“That process requires us to look into numerous records and testing data to have a better understanding of the complexities between various immunologic molecules between a given pair,” said Weimer, who is an associate professor in the Department of Pathology and Lab Medicine at the UNC School of Medicine.
A virtual crossmatch is a clinical risk assessment consultation that “virtually” determines if the donor’s immunologic profile matches a recipient’s immunologic profile well enough to move forward in a transplant. It’s a critical process full of complex medical data to make the best-educated decision for a transplant. But sometimes the answer can get convoluted.
“Like any laboratory test, there are false positives – when the assay performed is positive but the real or true answer is negative. That happens in crossmatching because of various disease states in a recipient, false reactivity that may occur due to the donor’s HLA [human leukocyte antigen] type or cells, for example,” said Weimer, who is also the associate director of the HLA laboratories at the UNC Medical Center.
The current process in trying to achieve a complex answer in a short period of time can potentially result in a backlog of organ waste if the assessment isn’t made quickly and accurately. Another issue is that organs do not survive very well outside the human body. Lab professionals have to act quickly because if the cold ischemia time increases in an organ, the quality of that organ for a particular recipient decreases.
Combining the expertise of math and immunobiology
What sparks innovation and advances in particular fields is when you intersect different disciplines. To improve the process for analyzing factors known to impact transplant outcomes, Weimer knew that the University of North Carolina had a diverse faculty that could help.
“It occurred to me years ago that crossmatching relies heavily on the use of numbers to make that virtual assessment,” he said. “That’s what led me to reach out to the mathematics department where I came across Dr. Katherine Newhall.”
As an applied mathematician in the UNC College of Arts and Sciences, Newhall’s work involves understanding scientific problems by mapping them to a mathematical model. When Weimer reached out, he used her expertise and techniques in an attempt to help predict recipient and donor immunologic compatibility.
“Dr. Weimer knew just enough about my field to know that math could help with this problem,” said Newhall, an associate professor in the UNC Department of Mathematics. “He wanted to know if I could collaborate to solve the problem of preemptively trying to decide if a patient and donor are compatible. So, we worked together to quantify that risk.”
A positive result means you are confident a patient is going to be incompatible for that donor. This is where Newhall’s stochastic background comes into play when dealing with these ideas of probability.
“Historically, potential transplant patients and donors get labeled as either compatible or incompatible. But, with my work, it’s not just black and white,” she said. “We can do more than just say yes or no. We can put these outcomes in a broader concept like: more likely or less likely. If you can reduce the number of times you can falsely label a person positive, then that means those people are getting off the waiting list and they’ll receive the organ.”
By combining their expertise of math and immunobiology, Newhall and Weimer created the Digital Alloimmune Risk Assessment, a new AI-driven approach to virtual crossmatching to improve organ transplantation.
From there, Weimer and Newhall partnered with Microsoft to harness the power of machine learning algorithms. The teams worked together to develop more sophisticated mathematics by combining math and biological data to teach computers how to solve the complex problem of matching donor and recipient.
“A lot of this Eric and I were doing by hand, like going through Excel sheets and changing the format and doing different things. Microsoft helped to automate all of that,” said Newhall. “With the help of machine learning and recognizing these patterns, we were able to create a more accurate risk assessment.”
Better matching. Better outcomes.
Five years in and researchers have already seen a dramatic increase in good results. This new software-based solution allows data analysis to be much more refined. It enhances the speed and understanding of a very complex biologic interaction. And this produces a more efficient evaluation of an essentially unlimited number of recipient-donor pair combinations.
Preliminary results on lung transplants show that the machine learning models can predict rejection before an organ is transplanted.
“If we can do these tests faster and they’re more accurate, then this platform can help change the standards of how these organs get allocated,” said Newhall.
Preliminary data show that in 2021, there were more than 41,000 organ transplants performed in the United States, according to the United Network of Organ Sharing. And with the help of this new tool created by Weimer and Newhall, that number could increase significantly in North Carolina and across the country.
“Overall, this solution streamlines the process by combining several different pieces of information in an easy to read and interpret manner,” said Weimer. “You can run a high number of patients against a given donor within minutes versus hours. This new process can be done seamlessly, so clinical providers and laboratory professionals all can be notified in part of that decision-making process.”
Today, Weimer and Newhall are working with partners to put these tools into the hands of transplant centers with the hopes of piloting with health systems as soon as possible.
“We’ve reached out to a number of transplant centers across the country because UNC is not alone in dealing with these digital and virtual crossmatches manually,” said Weimer. “This is where it can be really exciting because as people join onto this, we not only gain the ability to improve more patient outcomes, but we also gain access to working with a diversity of colleagues and patients to better understand transplant immunology in these donor-recipient pairs.”
At its core, the increase in speed, accuracy, precision, and reduction in false positives is what makes this new software not only a tool that reduces staff burdens, but it saves money and time – leading to an increase in saving lives.