The Myelodysplastic Syndromes Channel on VJHemOnc is an independent medical education platform, supported with funding from Geron (Silver). Supporters have no influence on the production of content. The levels of sponsorship listed are reflective of the amount of funding given.
EBMT 2025 | Developing a machine learning model for MDS prognosis & assessing the value of molecular data
Adrian Mosquera-Orgueira, MD, PhD, Santiago Clinic Hospital CHUS, Santiago, Spain, shares insights into some recent work exploring the development of a prognostic machine learning model in myelodysplastic syndromes (MDS), which achieved high accuracy in predicting overall survival and leukemic transformation. He highlights the potential for clinical models to be more accessible and effective than molecular data alone. This interview took place at the 51st Annual Meeting of the EBMT in Florence, Italy.
These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.
Transcript
This evidence arises from the Spanish myelofibrosis group. So these guys have collected thousands of patients in their registry with traditional clinical and cytogenetic data along with follow-up. And we developed a collaborative project for the Spanish group of MDS to tune a machine learning model that could take, again, leverage the machine learning tool to integrate all this information into a valuable prognostic tool that does not depend too much on next-generation sequencing data...
This evidence arises from the Spanish myelofibrosis group. So these guys have collected thousands of patients in their registry with traditional clinical and cytogenetic data along with follow-up. And we developed a collaborative project for the Spanish group of MDS to tune a machine learning model that could take, again, leverage the machine learning tool to integrate all this information into a valuable prognostic tool that does not depend too much on next-generation sequencing data.
So we actually developed this model in the Spanish group of MDS. And we observed very high accuracy in terms of concordance indexes, and that suggested to us that it could actually predict similarly to other molecular-enhanced models like the molecular IPSS. And nowadays, we have some data coming from South America and also some information, some data that will come from North America indicating that the clinical model is even slightly better than the molecular IPSS for overall survival and leukemic transformation prediction.
And this questions a little bit to what extent is molecular data needed for prognostication. I mean, maybe we better need to identify the subset of patients where molecular enhancement provides actually more than just the global set of patients, because we know that age is a very important parameter. And normally, this is not considered in the registries. The registries consider all kinds of patients. And then maybe we need to focus more on applying NGS for finding predictive patterns of response to differential response to therapies. Like what about AZA versus VEN plus AZA? You know, we know ASXL1 has an impact there. And that’s something that probably will shed more light rather than just raw prognostication, which can be done with more simple tools and with the benefit of enhancing accessibility to all.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
Read more...
