EBMT 2025 | Harnessing machine learning to predict outcomes post-alloSCT for patients with myelofibrosis

We have developed several projects in Spain within the Myelofibrosis group to risk stratify Myelofibrosis using machine learning. So these projects first have addressed the problem with clinical and traditional data. So we have collected a lot of clinical data from the registry, used machine learning to integrate this information and derive patient-specific scores.

Our first insights from this approach was that actually machine learning was capable of improving risk stratification to an unprecedented level compared even to some models that incorporate molecular data. So this is like a first evidence that precision medicine could be enhanced even with simple data through machine learning integration.

Then we integrated molecular data from also from the Spanish registry to enhance the identification of this tool. And we developed like a molecular enhanced tool that actually predicts better leukemia-free survival, so leukemic transformation and biological risk. However, we know that very few patients with myelofibrosis do undergo allogeneic stem cell transplantation because it’s difficult to find those candidates who are eligible for transplantation. And this is a complex choice. There is not much data or tools to provide decision making, accurate decision making.

So the idea was to develop an EBMT level project with taking leverage of the EBMT registry, which encompasses more than 5,000 patients with myelofibrosis, and identify through machine learning those high-risk patients who do perform very poorly after an allogeneic stem cell transplantation because we know this is probably the set of patients who should be very carefully considered to do an allogeneic stem cell transplantation because of high risk and little benefit.

So what we have developed with the EBMT is actually a tool that can identify 25 or 30 percent of patients who are actually high risk. These patients perform very poorly in survival and very similar to other tools that have been developed in the past for risk stratification after allogeneic hematopoietic stem cell transplantation in myelofibrosis but with a significant advantage that our machine learning tool is capable of identifying more than twice the number of patients who are high risk and this aligns better with reality and that’s why this is probably the first actionable machine learning model that the EBMT has developed for supporting decision making in allogeneic hematopoietic stem cell transplantation. It’s a great example of how artificial intelligence can be used to enhance decision making in medicine in general.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.