Researchers from the MUSC Hollings Cancer Center have developed a machine learning–based model that can identify patients at high risk of serious complications after stem cell and bone marrow transplantation months before clinical symptoms appear. The tool, called BIOPREVENT, focuses on predicting chronic graft-versus-host disease (GVHD) and transplant-related mortality. These are two major causes of long-term illness and death following transplantation.
For many patients, a stem cell or bone marrow transplant is life-saving. However, recovery does not end at hospital discharge. Chronic GVHD, in which immune cells from the donor attack the patient’s healthy tissues, can develop months later and affect organs such as the skin, eyes, mouth, joints and lungs. Once symptoms become apparent, significant and sometimes irreversible damage may already have occurred.
“By the time chronic GVHD is diagnosed, the disease process has often been unfolding for months, quietly hurting the body,” said Sophie Paczesny, MD, PhD, co-leader of the Cancer Biology and Immunology Research Program at Hollings. “We wanted to know whether we could detect warning signs earlier, before patients feel sick, and soon enough for clinicians to intervene.”
Combining biomarkers, clinical data and AI
To address this challenge, the research team applied machine learning to a large, multicenter dataset of 1,310 transplant recipients drawn from four major studies. Blood samples collected 90 to 100 days after transplant were analyzed for seven immune-related proteins associated with inflammation, immune activation and tissue damage. These biomarkers had been identified and validated in earlier research led by Paczesny.
The biomarker data were combined with nine routinely collected clinical variables, including patient age, transplant type, underlying disease and prior complications. These standardized clinical data were sourced from transplant registries maintained by the Center for International Blood and Marrow Transplant Research, ensuring consistent data quality across centers.
The researchers evaluated multiple machine learning approaches and found that a model based on Bayesian additive regression trees performed best. This approach became the foundation of BIOPREVENT.
Clear risk stratification
The results, published in the Journal of Clinical Investigation, show that models combining biomarkers with clinical data consistently outperformed models based on clinical information alone. Particularly when predicting transplant-related mortality. BIOPREVENT was also validated in an independent patient cohort, confirming its robustness beyond the original training data.
Importantly, the model could stratify patients into low- and high-risk groups with markedly different outcomes up to 18 months after transplant. The analysis also revealed that different biomarkers were associated with different outcomes, suggesting that chronic GVHD and transplant-related death are driven by overlapping but distinct biological processes.
From research to practice
To support broader use, the team translated BIOPREVENT into a free, web-based application. Clinicians can enter biomarker values and clinical details to receive individualized risk estimates over time.
“It was important to us that this not remain a theoretical model or a tool limited to a single institution,” Paczesny said. “Making BIOPREVENT freely available helps ensure that researchers and clinicians can test it, learn from it, and, ultimately, improve care for transplant patients.”
Precision transplant care
For now, BIOPREVENT is intended as a research and risk assessment tool, not a guide for treatment decisions. The next step will be clinical trials to determine whether acting on early risk signals, such as closer monitoring or preventive strategies, can improve long-term outcomes. “This isn’t about replacing clinical judgment,” Paczesny emphasized. “It’s about giving clinicians better information earlier so they can make more informed decisions.”
The study underscores the growing role of AI and data-driven models in enabling more personalized, proactive follow-up care for transplant patients, and potentially reducing the burden of one of transplant medicine’s most serious complications.
