2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Sunday, December 13, 2026
06:00 p.m. - 08:00 p.m.
MCCNO - Hall F
Technology and Procedures, Assays, Molecular Testing, Omics Technologies, Bioinformatics, Computational Biology, Emerging Technologies, Profiling, Treatment Considerations, Combination Therapy, Drug Development, Therapy Sequence, Biological Therapies, Antibody Therapy, Transplantation, Checkpoint Inhibitor, Immunotherapy, Non-Biological Therapies, Chemotherapy, Adverse Events, Artificial Intelligence (AI), Deep Learning, Machine Learning, Biological Processes, Apoptosis, Pathogenesis, Genomics, Hematopoiesis, Immune Mechanism, Metabolism, Molecular Biology, Multi-Systemic Interactions, Clinical Practice (Health Services and Quality), Practice Models, Value, Diseases, Myeloid Malignancies, Acute Myeloid Malignancies, Chronic Myeloid Malignancies, Measurable Residual Disease, Research, Fundamental Science, AML, APL, Translational Research, MDS, MPN, CMML, Clinical Research, Epidemiology, Health Outcomes Research, Health Disparities Research, Real-World Evidence, Survivorship, Study Population, Human, Adult
PAPER

3448 Transcriptomic profiling of leukemic blasts improves prognostic risk stratification beyond ELN2022 in NPM1-mutated acute myeloid leukemia

Vivian May Liu, MD, Guangrong Qin, PhD2, Megan Othus, PhD3, Jasmine Naru4, Jacob Appelbaum, MD/PhD5, Lan Beppu6, Thomas Chauncey, MD, PhD7, Harry Erba, MD, PhD8, John Godwin, MD9, Min Fang, MD, PhD, FACMG10, Stanley Lee, PhD11, Mary-Elizabeth Percival, MD12, Jerald Radich, MD13, Suravi Raychaudhuri, MD14, Cheryl Willman, MD15, Feinan Wu16, Soheil Meshinchi, MD, PhD17, Derek Stirewalt, MD18
1 Fred Hutchinson Cancer Center, Seattle, Washington, United States
2 Institute For Systems Biology, Seattle, Washington, United States
3 Fred Hutchinson Cancer Center, Seattle, Washington, United States
4 Fred Hutchinson Cancer Center, Seattle, Washington, United States
5 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
6 Fred Hutchinson Cancer Center, Seattle, Washington, United States
7 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States, VA Puget Sound Health Care System, Seattle, Washington, United States
8 Duke Cancer Center, Durham, North Carolina, United States
9 Providence Cancer Center, Portland, Oregon, United States
10 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
11 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
12 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
13 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
14 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
15 Mayo Clinic, Rochester, Minnesota, United States
16 Fred Hutchinson Cancer Center, Seattle, Washington, United States
17 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States
18 Fred Hutchinson Cancer Center, Seattle, Washington, United States, University of Washington Medical Center, Seattle, Washington, United States

NPM1-mutated (NPM1mut) AML is the most common molecular subtype of AML in adults. Risk stratification of patients at diagnosis relies primarily on genomic and cytogenetic features as defined by the European Leukemia Net 2022 guidelines (ELN22). The prognostic value of these models shows modest accuracy when predicting outcomes for specific patients, with concordance indices (C-indices) for overall survival (OS) and relapse-free survival (RFS) of 0.63 and 0.62, respectively (Termini et al. Haematologica 2023). We therefore sought to determine whether prognostic risk stratification in patients can be improved with transcriptomic profiling of gene expression in newly diagnosed NPM1mutAML.

Cryopreserved diagnostic bone marrow and/or peripheral blood specimens from NPM1mut AML patients were obtained from the SWOG Leukemia and the Fred Hutch/University of Washington Hematopoietic Disease Repositories (n=116). All patients received intensive induction and consolidation chemotherapy with curative intent, primarily on SWOG clinical trials (S9031, S9333, S0106, and S0112). Samples were enriched for viable leukemic blasts via flow cytometric sorting based on expression of CD34, CD117, or CD45 with low side scatter. Targeted next generation sequencing was used to identify genomic mutations, and fragment analysis was used to identify FLT3-ITD mutations. Bulk RNA sequencing (RNAseq) on enriched viable leukemic blasts was performed to quantify gene expression. RiboErase was utilized to deplete ribosomal RNA. Transcriptome libraries were generated using KAPA Stranded RNA-Seq Library Preparation Kit and sequenced in batches using either Illumina HiSeq 2500 or NovaSeq 6000 instruments. Transcripts were mapped, aligned to Hg38 genome build, and raw read counts were quantified using STAR. Transcript abundance was normalized using transcripts per million (TPM). Low abundance transcripts (log(TPM)<0 in at least 75% of samples) were excluded prior to analysis. Candidate prognostic transcripts were identified using univariate and multivariate Cox proportional hazards regression, adjusting for age and ELN22 risk group.

Leveraging the transcriptome data and clinical data, we developed prognostic models based on gene expression and clinical information. Nested leave-one-out-cross-validation (LOOCV) was used to identify the best parameters (penalizer and l1_ratio) in the CoxPHFitter function. Final models were trained with the best parameters for each training cycle and tested on the leave-one-out samples. C-indices were computed to evaluate the performances of models. We selected the most frequently selected parameters in the Nested LOOCV Training process using Elastic Regression models and applied it in the final model to predict OS. We grouped the patients by the predicted risk scores either by the best cutoff chosen by log-rank test statistic, or by quantiles. Significance was defined by log-rank test.

The median age of this NPM1mut cohort was 55 years (range 19-81), and the most common co-mutations were in DNMT3A and FLT3-ITD. Based on ELN22 criteria, 50 patients had favorable risk, 31 had intermediate risk, and 7 patients had adverse risk disease; the remaining patients (n=28) could not be classified due to lack of cytogenetic information. Multivariate analyses adjusted for age (as a continuous variable), age category (>60yrs or <= 60yrs) and ELN22 identified 14 genes with expression associated with OS. We constructed prognostic models for OS based on different combinations of feature sets including age, age category, ELN22, and gene expression signatures. The best performing prognostic model had a C-index of 0.77 and included the 14-gene expression signature, ELN22, age, and age category. This model outperforms the ELN22 based model (C-index: 0.52), and ELN22+Age model (C-index: 0.70). Using gene expression alone, our prognostic model shows similar performance with ELN22+Age model (C-index: 0.69). We are working to validate this 14-gene expression model in an independent cohort of NPM1mut AML patients from Fred Hutch.

Our novel prognostic model using a 14-gene expression signature, age, and ELN22 substantially improves prediction of OS beyond the established clinical risk factors in NPM1mut AML. Integrating RNAseq-based gene expression signatures with established risk models such as ELN22 may provide more accurate prognostic stratification of NPM1mut AML to improve clinical decision making.
Vivian May Liu, MD: No relevant conflicts of interest to declare.