The Acute Myeloid Leukemia Channel on VJHemOnc is an independent medical education platform, supported with funding from BMS (Silver), and through an educational grant from Jazz Pharmaceuticals. Supporters have no influence on the production of content. The levels of sponsorship listed are reflective of the amount of funding given.
ISAL 2025 | Implementing computational MRD assessment into clinical practice: challenges and solutions
Costa Bachas, PhD, Amsterdam University Medical Center, Amsterdam, Netherlands, comments on the challenges of implementing computational measurable residual disease (MRD) assessment in clinical practice. He highlights the need for thorough validations of algorithms to comply with regulatory requirements and medical equipment regulations. Dr Bachas emphasizes the importance of explainable statistical modeling to ensure reliable and interpretable results, particularly in stratifying therapy decisions for patients. This interview took place at the 19th International Symposium on Acute Leukemias (ISAL XIX) in Munich, Germany.
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
Implementing computational MRD assessment in clinical practice is of course a next step, but there are difficulties there. So actually algorithms have to comply with regulations of drug authorities, regulators like the FDA or EMA, and they have to comply with medical equipment regulations. So actually we need quite thorough validations of the algorithm. So we need to do that on multiple data sets, on independent data sets, acquired in different laboratories with different antibody panels to make sure that this really works well and gives reliable predictions...
Implementing computational MRD assessment in clinical practice is of course a next step, but there are difficulties there. So actually algorithms have to comply with regulations of drug authorities, regulators like the FDA or EMA, and they have to comply with medical equipment regulations. So actually we need quite thorough validations of the algorithm. So we need to do that on multiple data sets, on independent data sets, acquired in different laboratories with different antibody panels to make sure that this really works well and gives reliable predictions. So also we need to validate that prospectively. We need to implement it in a graphical interface that is accessible to clinicians. So we have quite a long way to go, I think. But an advantage of our approach is, like I said, that we really went for an explainable approach. So there are some other methods out there. We actually are going to publish a review, or I hope we will publish a review in Hemasphere. It’s now a publication that’s in the second round of revisions. And in that publication we explain why we think that our approach is one of the solutions to do this. That’s appropriate because for example we don’t use very black box machine learning algorithms. We really use explainable statistical modeling which is also very crucial because nowadays there are ways to interpret models. So machine learning models can indirectly be interpreted but that’s not the same as explaining a result for each individual patient. So it’s very important that if we predict a patient as MRD positive or MRD negative and stratify therapy according to that result which is happening based on MRD right now in the clinic. So either patients might go more quickly to a stem cell transplantation or might be kept from a stem cell transplantation. It’s very important that such a result is really reliable and explainable. So that’s what we aim to achieve. And we hope to validate that and go to a path of implementation.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
Read more...
