
Revolutionary Machine Learning Model Breaks Ground in Myelofibrosis Transplant Risk Prediction
2025-06-12
Author: Nur
A New Dawn in Myelofibrosis Treatment
Allogeneic hematopoietic cell transplantation (allo-HCT) has long been the only curative option for myelofibrosis, but with the rise of innovative treatments, pinpointing the right candidates for transplant has become increasingly complex. To tackle this challenge, a pioneering team from the European Society for Blood and Marrow Transplantation (EBMT) has unveiled a cutting-edge machine learning model designed specifically to stratify transplant risk for myelofibrosis patients.
EBMT’s Game-Changing Announcement
On March 27, 2025, the EBMT launched this groundbreaking tool that can accurately predict overall survival (OS) for myelofibrosis patients post-allo-HCT by analyzing various patient characteristics, including age, performance status, and comorbidity index. Best of all, this open-access resource is available for free online, enabling clinicians worldwide to utilize it.
Dr. Donal McLornan, an expert in myelofibrosis transplantation, expressed his enthusiasm, stating, "Despite nearly 20 years of experience, determining the ideal time and patient for transplant continues to be a daunting task, especially with the influx of new treatments that further complicate the decision-making process. This tool will be instrumental in identifying the right patients for transplant, alongside other crucial factors."
The Science Behind the Model
To develop this revolutionary tool, researchers meticulously analyzed data from adult patients diagnosed with primary or secondary myelofibrosis who underwent their first allo-HCT between 2005 and 2020 across EBMT centers. Using a random survival forests model, they incorporated 10 key variables: age, comorbidity index, performance status, blood blasts, hemoglobin levels, leukocytes, platelets, donor type, conditioning intensity, and graft-versus-host disease prophylaxis.
Data from 5,183 patients was meticulously examined, with 3,887 used for training the model and 1,296 for validation. A comparative analysis revealed the new machine learning approach outperformed traditional Cox regression-based scores and other existing models.
Significant Findings and Impacts
At a median follow-up of over 58 months, results showcased a median OS of 79.4 months in the training set and 73.7 months in the test cohort. The EBMT model demonstrated impressive precision in identifying high-risk patients, categorizing 25% as high risk with poor outcomes post-transplant. This figure starkly contrasts with only 10.1% identified by Cox scoring, aligning much more closely with the realities observed in clinical settings.
The model not only enhances risk stratification but also includes an interactive web-based calculator that provides clinicians with a visual representation of a patient’s risk score, complete with estimated OS and non-relapse mortality (NRM) rates.
A Tool for Informed Decision-Making
Dr. Juan Carlos Hernández-Boluda emphasized the significance of this tool, stating, "While we already have good transplant risk scoring systems, integrating toxicity insights alongside transplant outcomes is essential. This machine learning model will empower clinicians to make more informed decisions, ensuring patients receive the care they truly need, while also highlighting the inherent risks of transplantation."
With this innovative machine learning model, the landscape of myelofibrosis treatment is poised for transformation, offering hope not only for enhanced patient outcomes but also for a brighter future in stem cell transplantation.