2 Clarke Drive
Suite 100
Cranbury, NJ 08512
© 2025 MJH Life Sciences™ and OncLive - Clinical Oncology News, Cancer Expert Insights. All rights reserved.
Juan Carlos Hernández-Boluda, MD, PhD, discusses the development of an EBMT machine learning model for predicting allo-HCT risk in myelofibrosis.
Juan Carlos Hernández-Boluda, MD, PhD
An EBMT machine learning–based tool for identifying patients with myelofibrosis who are at high risk of allogeneic hematopoietic cell transplantation (allo-HCT)–related mortality offers a more comprehensive and accurate option compared with existing models, according to Juan Carlos Hernández-Boluda, MD, PhD.
“When you are considering taking a patient with myelofibrosis to transplant, you must first define if the patient is high risk,” Hernández-Boluda said in an interview with OncLive®. “We usually use standardized prognostic scoring systems. We currently have good scoring systems, but you also must consider the toxicity of the transplant; we don’t have very good tools [for this]. With this [machine learning] tool, we will have information on both parts.”
Hernández-Boluda was part of a team of investigators from the EBMT who designed a machine learning model based on data from adult patients with primary or secondary myelofibrosis who underwent their first allo-HCT between 2005 and 2020 in EBMT centers.1 Using a random survival forests model which considered 10 variables including patient age, donor type, and performance status, the EBMT model assigned 25% of patients as high risk for a poor outcome (approximately 35% non-relapse mortality and 40% overall mortality rates at 1 year) following transplantation compared with 10.1% using a Cox-derived score and 8.2% using the Center for International Blood and Marrow Transplant Research model. The open-access model was launched on March 27, 2025, and is available for free use online.2
In the interview, Hernández-Boluda, an associated physician in the Hematology and Medical Oncology Service at Hospital Clínico of Valencia, Spain, discussed the rationale for creating the machine learning model and what makes it unique, as well as highlights from a session he spoke at during the 51st Annual EBMT Meeting on future strategies to optimize outcomes following allo-HCT for myelofibrosis.
Hernández-Boluda: I’ve been managing patients with myelofibrosis for over 15 years and I’ve been using prognostic scoring systems to manage these patients with drugs, not transplant. Over the past [few] years, I’ve been working with Adrián Mosquera Orgueira, MD, PhD, on machine learning technology and we’ve seen that this methodology can produce reliable prognostic scoring systems outside of transplant. There are some limitations in the prognostic scoring systems that we have [for transplantation] and I thought it would be good to have a prognostic tool to [help us] decide which patients should be taken to transplant, conventional treatment, or clinical trials.
The technology is completely different and it’s more accurate than the available models. It’s also far more comprehensive because it is based on data from more than 5000 patients transplanted within EBMT centers and includes [data such as] graft source and methods of graft-vs-host disease [GVHD]. [These] outcomes [are] referred from nearly 300 centers of excellence in allo-HCT for [patients with] myelofibrosis as the standard of care. [The model has] accuracy and the outcomes represent the current strategies in this setting.
With this tool, we will have information on both the disease and the transplantation risk, and we can share this information with the patient. This scoring system also has a visual web application, and we can share this image with the patient if they are interested. [This tool will allow] us to make a more informed decision which is critical for [the patient’s disease] management. Transplantation is a curative treatment, but it also poses a lot of danger in terms of the life of the patient. I believe combining what we had before with this tool is going to help us to make that decision [more effectively].
The first aim for me was to give the up-to-date, proper conditioning [regimen] for GVHD prophylaxis, depending on the patient and disease characteristics, as well as donor source. The session had 2 parts. It was an educational session trying to [outline] what we are doing now. But there [were] a couple of interesting things I also mentioned in GVHD prophylaxis that we don’t have enough research for so far, [such as] the use of post-response cyclophosphamide or JAK inhibitors.
Related Content: