-Author name in bold denotes the presenting author
-Asterisk * with author name denotes a Non-ASH member
Clinically Relevant Abstract denotes an abstract that is clinically relevant.

PhD Trainee denotes that this is a recommended PHD Trainee Session.

Ticketed Session denotes that this is a ticketed session.

4307 MDS-AML Aggregative Risk Classification System (MARCS): A Novel, Enhanced Risk Stratification System Derived from the Combined Unsupervised Analysis of 7,480 MDS and AML Patients

Program: Oral and Poster Abstracts
Session: 618. Acute Myeloid Leukemias: Biomarkers and Molecular Markers in Diagnosis and Prognosis: Poster III
Hematology Disease Topics & Pathways:
Research, Artificial intelligence (AI), Translational Research, Genomics, Bioinformatics, Computational biology, Biological Processes, Technology and Procedures, Machine learning, Omics technologies
Monday, December 9, 2024, 6:00 PM-8:00 PM

Marco Roncador1,2,3*, Fritz Bayer, PhD2,4*, Jeremy W. Deuel, MD, PhD3*, Jack Kuipers, PhD2*, Markus G. Manz, MD3, Steffen Boettcher, MD3*, Niko Beerenwinkel, PhD2,4* and Stefan Balabanov, MD, PhD3*

1Department of Biosystems Science and Engineering, ETH Zurich, Zurich, Switzerland
2SIB Swiss Institute of Bioinformatics, Basel, Switzerland
3Department of Medical Oncology and Hematology, University Hospital Zurich and University of Zurich, Zurich, Switzerland
4Department of Biosystems Science and Engineering, ETH Zurich, Basel, Switzerland

Myelodysplastic neoplasia (MDS) and acute myeloid leukemia (AML) share common clinical and genetic features. With increasing knowledge on genomic drivers, morphology-based definitions are being increasingly replaced by molecular definitions in the classification systems of both AML and MDS. Due to similarities of genetic drivers, a growing body of evidence questions the dichotomy between these two diseases, but rather suggests that they are evolving entities within a disease continuum.

We studied a large cohort of 7,480 patients encompassing the majority of MDS (n=3729) and AML (n=3751) subtypes acquired between 2008 and 2019. For each subject, data on blood parameters (hemoglobin level, white blood cell count, thrombocytes, and bone marrow blasts), the mutational status of 32 genes (ASXL1, BCOR, CBL, CUX1, DNMT3A, ETV6, EZH2, GATA2, GNAS, IDH1, JAK2, KIT, KRAS, MPL, NF1, NPM1, PHF6, PTPN11, RAD21, RUNX1, SF3B1, SRSF2, STAG2, TET2, TP53, U2AF1, WT1, ZRSR2, IDH2, NRAS, CEBPA, FLT3), and selected cytogenetic data (trisomy 8, loss of chromosome 17, del(5q) or monosomy 5, del(7q) or monosomy 7) were available. We analyzed the data using a novel covariate-aware clustering method (CANClust) and developed the novel disease stratification system MARCS (MDS-AML Aggregative Risk Classification System). This new classification system was validated on an independent cohort of 1,035 MDS (n=489) and AML (n=546) patients, confirming the results obtained from the original cohort.

We categorized patients into nine risk groups, which coherently recapitulate the accumulated genetic knowledge on MDS and AML subclasses, including the standalone behavior of NPM1-mutated AML and an ultra-high-risk patient group with TP53 mutation. The new system also aligns with known good prognostic markers such as SF3B1 mutation or del(5q). The MARCS-driven stratification predicted disease evolution and patient outcomes more effectively than the 2022 European LeukemiaNet genetic risk stratification (ELN2022) and the Molecular International Prognostic Scoring System (IPSS-M), even when both scores are used together (likelihood ratio (LR) = 501.8; Wilks test p-value < 10-4, n = 7,480, degrees of freedom (df) = 19). The MARCS integrative clinico-genomic stratification is particularly effective in predicting outcomes for patients within the novel International Consensus Classification (ICC) MDS/AML subclass in comparison with the IPSS-M score (Log-rank p-value <0.0001). Applying the same analytical process to learn risk groups in AML and MDS patients separately and combining the two clusterings afterward, yielded a worse overall prediction than deriving the risk categories directly from the combined dataset of AML and MDS patients as performed for MARCS (LR = 444.5 vs 502.6, Wilks test p-value < 10-4, df = 11 vs 10).

In conclusion, using CANClust the MARCS classification system dynamically combines the mutational landscape with clinical parameters to regroup MDS and AML patients into nine novel risk groups, which predict the actual clinical course more accurately, transcending the initial MDS-AML separation and outperforming the ELN2022 and IPSS-M scores. The definition of disease type based on blast percentage is additionally questioned by our analysis, as many clusters include patients with blast counts both above and below the diagnostic cutoff of 10-20%. MARCS is particularly effective in predicting the outcomes of high-risk patients and may serve as a guide in deciding the treatment for aggressive diseases. The novel classification is accessible through an open-access website (https://marcs.ethz.ch/).

Disclosures: Boettcher: Pfizer: Consultancy; Servier: Consultancy; Astellas: Consultancy.

*signifies non-member of ASH