Share on PinterestAn AI tool may be able to predict GVHD risk, prompting earlier treatment to prevent complications. Image credit: Victor Bordera/Stocksy
- An AI-based tool may be able to predict the risk of developing chronic graft-versus-host disease (GVHD) and transplant-related death after stem cell or bone marrow transplant.
- Combining biomarkers with clinical factors, the AI tool predicted outcomes more accurately than clinical data alone, particularly for transplant-related mortality.
- The tool arranged patients into low- and high-risk groups, with clear differences in outcomes up to 18 months post-transplant, and was validated in an independent patient cohort.
- The machine learning model is available as a free, web-based application to support risk assessment and research.
Stem cell and bone marrow transplants are procedures that replace diseased, damaged, or destroyed blood-forming cells with healthy tissue. They are a common treatment for leukemia, lymphoma, and blood disorders.
These procedures involve harvesting cells from a donor (allogenic) or using the patient’s own cells (autologous). For many people, transplantation can be lifesaving. However, recovery does not end after leaving the hospital.
Potential complications can result in treatment-related mortality, typically driven by GVHD. Although advances in transplant care have improved survival rates, GVHD is the leading cause of late morbidity and mortality after an allogenic stem cell transplant.
It is difficult to predict who will experience GVHD and who will not. However, evidence suggests that between half to a third of all people who have an allogeneic transplant develop some symptoms of GvHD.
It can occur shortly after the transplant, known as acute GVHD, or can arise months after the transplant, called chronic GVHD (cGCHD).
Preventing GVHD can be challenging, as this typically involves balancing immune suppression to prevent GVHD without increasing infection risk and preventing adverse reactions to these treatments.
A new study, published in the Journal of Clinical Investigation, describes a machine-learning model that estimates a patient’s risk of developing cGVHD and dying from transplant-related causes before symptoms appear.
Researchers suggest the tool could give clinicians an early warning and open a window for closer monitoring or preventive strategies.
Turning blood tests into predictive insights
To develop the AI tool, known as BIOPREVENT, a collaborative team, led by researchers from the Medical University of South Carolina (MUSC) Hollings Cancer Center, analyzed data from 1,310 transplant recipients enrolled in four large, multicenter studies.
The team focused on blood samples collected 90 to 100 days after transplant. Lead study author Sophie Paczesny, MD, PhD, told Medical News Today that this window is a critical period when patients may feel well, but underlying immune activity may already be setting the stage for complications.
“Disease does not start when symptoms appear — it starts silently. We hypothesized that around day 90 to 100 there is a subclinical phase of cGVHD that can be detected biologically before it becomes clinically apparent,” Paczesny noted.
“Our data suggest that biomarker-informed machine learning can identify risk approximately 2 to 8 months before formal diagnosis — creating a window of opportunity for earlier action.”
— Sophie Paczesny
Previous work led by this researcher, who serves as co-leader of the Cancer Biology and Immunology Program for the Medical University of South Carolina Hollings Cancer Center, had identified and validated seven immune-related proteins linked to inflammation, immune activation, immune regulation, and tissue injury.
The researchers measured these seven biomarkers, combined with nine clinical factors, including age, transplant type, primary disease, and prior complications.
The researchers tested multiple machine-learning approaches to determine which could best predict outcomes. The strongest-performing model used a statistical method known as Bayesian additive regression trees, which became the foundation for BIOPREVENT.
Paczesny told MNT that “cGVHD remains one of the most debilitating complications after hematopoietic cell transplantation.”
“Our study shows that a machine learning model using blood biomarkers at three months post-transplant can predict who is at risk months before symptoms appear—opening the door to earlier, potentially preemptive intervention,” she added.
More accurate than clinical data alone
The study results showed that combining biomarker data with clinical information significantly improved the ability to predict transplant-related mortality compared with using clinical data alone.
Importantly, the team validated the AI model in an independent group of transplant recipients, confirming that the tool could reliably predict risk beyond the patients used to build it.
Additionally, BIOPREVENT successfully categorized individuals into low- and high-risk groups, with clear differences in outcomes up to 18 months after transplant.
The findings also suggest that cGVHD and transplant-related death may be driven by partly distinct biological processes.
Certain biomarkers were more strongly associated with the risk of death after transplant, while others were better predictors of who would later develop cGVHD.
From research model to web-based tool
To help encourage broader use, the researchers made BIOPREVENT freely available as a web-based application. Clinicians can enter a patient’s clinical characteristics and biomarker values to receive personalized risk estimates over time.
“Once entered into the app, the model generates an individualized risk score. This enables more precise monitoring and earlier clinical decision-making—bringing us closer to personalized care,” Paczesny told MNT.
For now, the researchers note that BIOPREVENT is designed to support risk assessment and research rather than to directly guide treatment decisions.
The next step will involve clinical trials to determine whether acting on early risk signals, such as increasing monitoring or offering preventive therapies to high-risk patients, can improve long-term outcomes.
A move toward precision transplant care
The study reflects a broader shift toward precision medicine in transplant care, where follow-up and treatment strategies are tailored to a patient’s individual risk profile. The AI tool may be able to provide clinicians with better information earlier, so they can make more informed decisions.
“Early prediction shifts the paradigm from reactive to preemptive care,” Paczesny explained.
“For patients, this could mean closer, personalized monitoring if they are high risk; earlier therapeutic intervention at the first subtle signs; and ultimately, enrollment in preemptive trials designed specifically for high-risk individuals,” the researcher told us.
“My priority is to collaborate across institutions to launch a preemptive trial in which high-risk patients receive one of the currently approved non-steroid agents before irreversible damage occurs,” she added.
Although further validation is necessary before BIOPREVENT becomes part of routine care, this approach could represent a promising step toward reducing one of transplant medicine’s most serious complications.
