My name is Lisa Pleyer. I’m a full professor of hematology in Salzburg, Austria. My talk this year at ASH focused on using advanced techniques, including machine learning, to evaluate whether bone marrow assessments are still necessary for patients in 2025.
We analyzed a large database of over 4,000 patients, which included highly detailed data from real-world patients as well as two phase three clinical trials. Among these, 3,000 patients had MDS, CMML, or AML, and we examined more than 200,000 data points, including detailed blood counts and other measures that do not require bone marrow.
Why did we do this? Bone marrow evaluations are painful procedures, and it usually takes about two weeks to get histology results. In the real world, patients only receive bone marrow assessments about 50% of the time during follow-up. Even in clinical trials, only about 66% of patients get them. Yet, current response criteria rely heavily on bone marrow results, and if a patient doesn’t have one, they’re automatically counted as non-responders.
Our goal was to find alternative response measures that don’t require bone marrow. We input all 200,000 data points into a complex AI model and analyzed outcomes across the whole cohort and by disease subgroups—AML, MDS, and other patients—as well as by treatment type, including HMA-based therapies, chemotherapy, or other therapies.
The results were striking. In all subsets, the AI model showed that rich peripheral blood data alone was sufficient for outcome prediction. When we removed all bone marrow-related variables, the model’s predictive accuracy did not decrease.
In summary, this work suggests that bone marrow evaluations are not as critical as previously thought when high-quality peripheral blood data is available. Our next step is to identify the most important parameters and cutoff values to define a response type capable of distinguishing patients with different prognosis trajectories—without the need for a bone marrow evaluation.
Importantly, we validated these findings using both real-world and randomized phase three clinical trial datasets, confirming the robustness of our approach.
Thank you for your attention. If our videos have helped you in any way and you’re able to, please consider making a donation to help us continue this important work. Your gift will go three times as far when we reach $500,000 by the end of the year. Every contribution, big or small, makes a difference, and we’re deeply grateful for your support.
HealthTree • December 23, 2025