Risk-Adapted Donor Selection in Patients with Acute Myeloid Leukemia
Related Experiment Video
Updated: Nov 8, 2026
Murine Model of Leukemia Relapse to Induction Chemotherapy for Acute Lymphoblastic Leukemia
Published on: October 17, 2025
Xavier Cheng-Hong Tsai1, Jia-Hau Liu2, Liang-Tsai Hsiao3
1Division of Hematology, Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan; Department of Hematological Oncology, National Taiwan University Cancer Center, Taipei, Taiwan.
Background:
The optimal donor for adults with acute myeloid leukemia (AML) who lack a matched sibling donor (MSD) remains uncertain in Asian practice, where antithymocyte globulin (ATG)-based graft-versus-host disease (GvHD) prophylaxis is common.
Objectives:
To compare outcomes after MSD, matched unrelated donor (MUD), mismatched unrelated donor (MMUD), and haploidentical donor (HID) transplantation, and to test whether disease risk modifies the donor effect.
Study Design:
We analyzed 1,963 adults with AML who underwent first allogeneic hematopoietic stem cell transplantation (HSCT) in the Taiwan Blood and Marrow Transplantation Registry between 2009 and 2025. We applied a doubly robust strategy combining multinomial propensity-score inverse probability of treatment weighting (IPTW) with multivariable Cox and Fine-Gray regression. GvHD was evaluated as a mediator by landmark and causal mediation analyses.
Results:
Using MUD as reference, all three other donor types had inferior overall survival (OS), reaching significance for HID (hazard ratio [HR] 1.34, p=0.022) and MMUD (HR 1.31, p=0.026) but not for MSD (HR 1.18, p=0.131). Non-relapse mortality (NRM) was higher after HID (HR 1.61, p=0.011), with a trend after MMUD (HR 1.38, p=0.079), and GvHD-free, relapse-free survival (GRFS) was inferior after MSD (HR 1.40, p<0.001) and HID (HR 1.31, p=0.008). Relapse did not differ by donor type, indicating that the survival advantage of MUD operates through lower NRM. Causal mediation analysis did not attribute the MUD-HID survival difference to GvHD. Disease risk modified the donor effect on OS (interaction p=0.028). In high/very-high-risk disease, MUD was superior to MSD, MMUD, and HID (HR 1.44, 1.78, and 1.74, respectively), with 3-year OS of 42.7% after MUD and 28.5% after other donors. In low/intermediate-risk disease, OS did not differ by donor type.
Conclusions:
MUD was associated with better OS than alternative donors (MMUD and HID) through lower NRM, and its survival benefit was largest in high-risk disease. Donor selection in AML can be risk-adapted: MUD could be prioritized in high-risk disease, whereas GRFS should guide donor choice in lower-risk disease, where OS is similar across donor types.
