Stem cell transplantation: ai to predict severe complications

Researchers at the MUSC Hollings Cancer Center sought to identify these early, invisible warning signals. Their objective was to predict the future risk of cGVHD and transplant-related mortality before any clinical manifestation.

“By the time chronic GVHD is diagnosed, the disease has often been progressing for months, silently affecting the body,” explained Dr. Sophie Paczesny, co-leader of the Cancer Biology and Immunology Research Program at Hollings and co-author of the study, in a press release. “We wanted to know whether it was possible to detect early warning signs before patients even begin to feel ill.” The findings were published on February 16 in the Journal of Clinical Investigation.





AI, biomarkers, and clinical data



The study analyzed 1,310 patients from four multicenter studies, all of whom had undergone stem cell or bone marrow transplantation. Blood samples collected between 90 and 100 days post-transplant were analyzed to measure seven immune-related proteins associated with inflammation, immune activation, and tissue remodeling.

These biomarkers were combined with nine clinical variables—such as age, transplant type, underlying disease, and prior complications—drawn from the Center for International Blood and Marrow Transplant Research registry.

Several machine learning models were tested. The best-performing model, based on Bayesian additive regression trees, served as the foundation for the BIOPREVENT tool, designed to estimate future risk of cGVHD and transplant-related mortality.




A signal before disease onset



Models integrating both biomarkers and clinical data consistently outperformed clinical-only models, particularly in predicting transplant-related mortality. The tool was subsequently validated in an independent cohort, confirming its robustness.

BIOPREVENT stratifies patients into low- and high-risk groups, with clearly distinct clinical trajectories up to 18 months post-transplant. Notably, some biomarkers were specifically associated with mortality, while others were more strongly predictive of cGVHD onset, suggesting partially distinct biological mechanisms.

  To facilitate adoption, BIOPREVENT was developed into a free web-based application, allowing clinicians to input biological and clinical data to obtain a personalized risk estimate. “It was important to us that this project not remain a theoretical model,” Dr. Paczesny emphasized. “Making BIOPREVENT freely accessible allows researchers and clinicians to test it and improve patient management.” At present, the tool is intended for research and risk assessment purposes and does not directly guide therapeutic decisions.  




Toward precision transplantation


This study represents a key step toward precision medicine in transplantation, based on integrating biological and clinical data. Clinical trials will be necessary to determine whether enhanced surveillance or early interventions in high-risk patients can truly improve outcomes.

“It is not about replacing clinical judgment,” Dr. Paczesny concluded, “but about providing better information earlier so clinicians can make more informed decisions.” Ultimately, BIOPREVENT could transform post-transplant follow-up—shifting from a reactive approach to complications toward a proactive approach to risk.



                     Read next: Will biomarkers save patients with refractory testicular germ cell tumors?








About the author – Elodie Vaz 
Health journalist, graduated from the CFPJ in 2023 Élodie explores the marks that illnesses leave on bodies and, more broadly, on human life. A state-registered nurse since 2010, she spent twelve years at patients’ bedsides before trading her stethoscope for a notebook. She now examines the connections between environment and health, convinced that the vitality of life cannot be reduced to that of humans alone.