Dr McLornan on the Future of an EBMT Machine Learning Model for Identifying and Stratifying Transplant Risk in Myelofibrosis

“I’ve worked in transplantation in [patients with] myelofibrosis for almost 2 decades and, despite that experience, choosing the right time and the right patient to bring forward for transplant is still challenging.”

Donal McLornan, MBBCh, MRCP, PhD, FRCPath, consultant, Hematology and Stem Cell Transplantation, University College London Hospitals NHS Foundation Trust, discusses future research directions for an EBMT machine learning model for identifying and stratifying transplant risk in myelofibrosis.

On March 27, 2025, a team from the European Society for Blood and Marrow Transplantation (EBMT) announced that a machine learning model they had developed for identifying and stratifying transplant risk for patients with myelofibrosis outperformed standard statistical models. The machine learning approach was trained using records of 3887 patients with myelofibrosis who received their allogeneic-hematopoietic cell transplantation between 2005 and 2020. Data from 1296 patients were used to assess and validate the model.

The open-access tool outperformed standard models in terms of accuracy, identifying a subset of patients with high-risk disease who had a 40% chance of dying within 1 year of transplant and a non-relapse mortality rate of approximately 35%. Moreover, the model identified 25% of patients as being a part of this group at a high-risk for a poor outcome after transplantation, compared with 10.1% via a Cox-derived score and 8.2% using the Center for International Blood and Marrow Transplant Research model.

Despite his nearly 20 years of experience, selecting patients with myelofibrosis who are good candidates for transplantation remains a challenge, partially due to patient interest in clinical trials and new agents, McLornan said. Additionally, the integration of these new approaches into the transplantation algorithm is not fully understood, he continued. The machine learning model, along with several other factors, will aid clinicians in understanding which patients should be candidates for transplant, he noted.

McLornan encourages investigators to use the tool in order to gain familiarity with how it functions. In the future, the EBMT team hopes to integrate the mutational profile of the patient, as well as the prior agents they have received with sequencing information into the machine learning tool, he continued. This information would further refine the model, increasing its usefulness, he concluded.