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

Paper Details

Abstracts, authors, affiliations, and disclosures.

Monday, December 14, 2026
02:45 p.m. - 04:15 p.m.
MCCNO - 353-355
Technology and Procedures, Assays, Molecular Testing, Treatment Considerations, Women’s Health, Workforce, Artificial Intelligence (AI), Clinical Practice (Health Services and Quality), Patient Safety, Quality Improvement, Diseases, Bone Marrow Failure Syndromes, Myeloid Malignancies, Acute Myeloid Malignancies, Measurable Residual Disease, Research, AML, Clinical Research, Study Population, Human
PAPER

948 Alloheme maintains relapse-prediction performance across clinical risk groups and molecular profiles following allogeneic HCT for AML and MDS

Nelli Bejanyan, MD, Ronald Sobecks, Professor of Medicine2, Stefan Ciurea, MD3, Piyanuch Kongtim, MD, PhD4, Sagar S Patel, MD5, Iskra Pusic, MD MSCI6, Vamsi Kota, MD7, Ran Reshef, MD, MSc8, Nabeel Rajeh, MD9, Jing Shi, PhD10, William Lento, PhD11, Nishant Dwivedi, MD, PhD12, Arpita Gandhi, M.D., M.S.13, Corey Cutler, MD,FRCPC,MPH14, Monzr M. M Al Malki, MD15
1 Moffitt Cancer Center, Tampa, Florida, United States
2 Taussig Cancer Institute, Cleveland Clinic, Cleveland, Ohio, United States
3 Chao Family Comprehensive Cancer Center, University of California, irvine, California, United States
4 Chao Family Comprehensive Cancer Center, University of California, irvine, California, United States
5 . Huntsman Cancer Institute, University of Utah, Salt Lake City, Utah, United States
6 Washington University School of Medicine, st louis, Missouri, United States
7 Georgia Cancer Center, Augusta University, Augusta, Georgia, United States
8 Columbia University Herbert Irving Comprehensive Cancer Center, New York, New York, United States
9 Saint Louis University School of Medicine, st louis, Missouri, United States
10 CareDx, Brisbane, California, United States
11 CareDx, Brisbane, California, United States
12 CareDx, Brisbane, California, United States
13 Knight Cancer Institute, Oregon Health and Science University, Portland, Oregon, United States
14 Dana Farber Cancer Institute, boston, Massachusetts, United States
15 City of Hope National Medical Center, Duarte, California, United States

Introduction   

AlloHeme is a peripheral blood–based, AI-powered test, clinically validated in the ACROBAT multi-center study (NCT04635384) to predict relapse after allogeneic hematopoietic cell transplantation (HCT) for acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS). We previously reported strong AlloHeme performance for relapse detection in the overall cohort: sensitivity 85%, specificity 92%, PPV 79%, NPV 95%, with a median lead time of 41 days before relapse. Here, we report AlloHeme performance across major disease subgroups and the spectrum of pre-HCT molecular profiles.   

Methods   

Adult patients undergoing a first allogeneic HCT for AML or MDS at 11 transplant centers were serially tested with AlloHeme. Standard-of-care (SOC) chimerism (STR-PCR, qPCR, or NGS) and MRD tests (bone marrow multiparameter flow cytometry, MFC-MRD, or molecular MRD) were performed per SOC surveillance; chimerism positivity used increasing mixed chimerism (IMC) at 1% or 5% per site technology.  The AlloHeme AI algorithm integrates multi-analyte inputs and serial whole-blood and myeloid fraction chimerism measurements with time post-transplant to generate test results. Performance was assessed in the cohort using patient-level leave-one-out cross-validation. 

AlloHeme testing was performed post-HCT using predefined sample collection timepoints after transplant. Longitudinal peripheral blood NGS chimerisms were integrated into a supervised machine-learning classifier. AlloHeme predictive performance was evaluated in clinical subgroups including disease type (AML, MDS), donor type, conditioning intensity, disease risk index (DRI), pre-transplant MRD status, graft source, and GVHD prophylaxis. For each pre-HCT molecular MRD mutation, AlloHeme performance was compared between mutation-positive and mutation-negative patients. Decision curve analysis compared the net benefit of AlloHeme versus SOC-guided surveillance. Relapses were determined by treating physicians using the Center for International Blood and Marrow Transplant Research (CIBMTR) criteria. Analyses used R 4.4.2 (subgroup differences by Fisher exact test; two-sided P<0.05 was significant).   

Results   

Overall, after censoring early study events (4 relapse, 4 death, 4 withdrawal before day 49 post-HCT), 198 subjects (143 AML, 55 MDS) were included, with 40 relapses and 26 non-relapse deaths. Among the 40 relapses, AlloHeme was positive in 33 cases ahead of relapse determination resulting in a predictive test performance of 83% sensitivity (33/40), 92% specificity, 79% PPV, 94% NPV, and an AUC of 0.87. By comparison, bone marrow MFC-MRD and molecular MRD were positive before disease relapse in only 5/39 (12%) and 4/39 cases (10%), respectively.  In a decision curve analysis, AlloHeme provided greater net benefit than SOC surveillance at all thresholds examined (33 vs 15 true positives, 9 vs 20 false positives).   AlloHeme performance was consistent across all clinical subgroups: sensitivity 75–93%, specificity 88–100%, AUC 0.83–0.92, with no significant between-subgroup difference (all P>0.05). 

Analysis of mutation status identified 7 frequent pre-HCT molecular MRD mutations (FLT3-ITD/TKD, DNMT3A, TET2, RUNX1, SRSF2, CEBPA and TP53), and the comparison of mutation-positive vs mutation-negative patients showed no significant difference in sensitivity, specificity, PPV, or NPV (all P>0.05). AlloHeme performance did not differ among DRI-stratified subjects (low N=39, intermediate N=77, high N=82):  sensitivity 100% (5/5), 73% (8/11), and 83% (20/24); specificity 92%, 96%, and 89% (heterogeneity P>0.05 for both).  

Conclusions   

In this multi-center cohort, AlloHeme had reproducible relapse-prediction performance across disease risk categories and pre-transplant molecular profiles without significant heterogeneity in the populations. AlloHeme performs consistently better than standard MRD testing for relapse prediction, and the data supports its broad applicability post-HCT.   

Nelli Bejanyan, MD: Orca bio - Consultancy (Includes expert testimony); Tscan Therapeutics - Consultancy (Includes expert testimony); Incyte, Inc. and Subsidiaries - Consultancy (Includes expert testimony); Medexus Pharma - Consultancy (Includes expert testimony); CareDx - Consultancy (Includes expert testimony)