Vascular Comorbidities Double Mortality Risk in Haploidentical Transplants

A machine-learning analysis identifies a vascular penalty and a three-factor tool to optimize donor selection in allogeneic HCT.

For Doctors in a Hurry

Clinicians lack precise tools to balance relapse risks against non-relapse mortality when selecting donors for hematopoietic cell transplantation.
The researchers analyzed 1,713 discovery patients and 6,829 replication patients to model individualized outcomes using machine learning algorithms.
Patients with vascular comorbidities faced a sub-distribution hazard ratio of 2.09 for mortality when receiving haploidentical transplants.
The authors concluded that relapse risk is driven by disease biology while non-relapse mortality depends on physiological reserve.
A new three-factor nomogram may help identify older patients who achieve equivalent disease-free survival with matched haploidentical donors.

Refining Donor Selection in Allogeneic Transplantation

Allogeneic hematopoietic cell transplantation remains a primary curative strategy for high-risk hematologic malignancies, yet its utility is frequently constrained by the competing risks of disease relapse and treatment-related toxicity [1, 2]. While human leukocyte antigen matching (the use of genetic protein markers to ensure donor-recipient compatibility) has traditionally dictated donor selection, the rising use of haploidentical donors (half-matched family members) has introduced complex trade-offs between potent graft-versus-tumor effects and higher rates of non-relapse mortality [3, 4]. Current clinical practice relies heavily on standardized conditioning regimens and graft-versus-host disease prophylaxis to mitigate these risks, but individual patient outcomes remain difficult to predict [5, 6]. For example, a meta-analysis of 975 patients demonstrated that the busulfan and fludarabine regimen was associated with a significantly lower one-year non-relapse mortality compared to busulfan and cyclophosphamide (risk ratio 0.49; 95% confidence interval 0.31 to 0.79) [1]. As the field moves toward more personalized transplant protocols, understanding the specific physiological factors that drive transplant-related mortality is essential for optimizing donor choice [7]. A new study now offers fresh insights into the biological factors that decouple treatment efficacy from systemic toxicity.

Decoupling Relapse from Physiological Toxicity

To investigate the drivers of transplant outcomes, the researchers utilized a discovery cohort of 1,713 adult patients from MD Anderson Cancer Center and validated these findings using registry replication cohorts totaling 6,829 patients. The study employed a double machine-learning framework known as Causal Survival Forests (a statistical method used to estimate how treatment effects vary across individuals based on their specific characteristics) to determine individualized treatment effects. To complement this, the authors used Random Survival Forests (an ensemble learning method that handles complex, non-linear interactions in survival data) to define the absolute prognostic risk for each patient. The analysis demonstrated a clear biological uncoupling of relapse and non-relapse mortality, suggesting that these two primary causes of treatment failure are influenced by distinct clinical factors. The researchers found that relapse was driven primarily by disease biology, such as the specific cytogenetic characteristics and aggressiveness of the underlying malignancy. Conversely, non-relapse mortality was governed by the patient's physiological reserve, which refers to the individual's baseline systemic health and their capacity to withstand the intensive conditioning and immunological stress of the transplant process. For the practicing clinician, this distinction emphasizes that while disease markers predict relapse, the patient's underlying comorbidities are the primary determinants of transplant-related toxicity.

The Vascular Penalty in Haploidentical HCT

While haploidentical hematopoietic cell transplantation offers potent graft-versus-tumor effects that are critical for achieving long-term remission, its clinical utility has been historically limited by higher non-relapse mortality compared to matched unrelated donors or matched related donors. This increased risk of treatment-related death often forces clinicians to balance the necessity of rapid donor availability against the potential for severe post-transplant complications. The researchers sought to identify the specific physiological factors that drive this mortality gap, discovering that the patient's baseline vascular health plays a disproportionate role in outcomes following haploidentical procedures. The analysis revealed that cardiovascular or cerebrovascular comorbidities acted as selective amplifiers of haploidentical transplant toxicity, creating what the authors termed a "vascular penalty." This finding indicates that pre-existing conditions such as coronary artery disease or a history of stroke do not merely increase general risk but specifically interact with the haploidentical transplant platform to worsen outcomes. In the study's bootstrapped model (a statistical technique that uses repeated resampling of the data to ensure the stability and reliability of the findings), this vascular penalty doubled the mortality risk for patients undergoing haploidentical transplantation. The statistical impact of these comorbidities was quantified by a sub-distribution hazard ratio of 2.09 (95% confidence interval 1.26 to 3.35), a metric used to estimate the probability of an event occurring when there are competing risks, such as relapse and death. For the practicing clinician, these data suggest that the presence of vascular disease is a primary determinant of whether a patient can tolerate the immunological stress of a haploidentical graft.

A Three-Factor Tool for Donor Optimization

To translate these findings into clinical practice, the researchers developed a three-factor nomogram (a graphical calculating device used for predicting clinical outcomes) designed to identify what they termed the Optimized Haploidentical phenotype. This tool allows clinicians to determine which patients are most likely to benefit from a haploidentical transplant by integrating three specific variables: recipient age, the presence of vascular comorbidity, and B-leader matching status. The latter factor, B-leader matching status, refers to a specific genetic matching criterion for the HLA-B leader peptide that influences how Natural Killer cells interact with the graft. By synthesizing these physiological and genetic markers, the nomogram provides a structured method for risk stratification, moving beyond traditional donor-recipient matching to include the recipient's baseline health and specific immunogenetic compatibility. The clinical relevance of identifying an Optimized Haploidentical phenotype lies in its ability to pinpoint a subset of patients where the risks of non-relapse mortality are sufficiently mitigated. However, the study's exploratory multivariable modeling provided a clear hierarchy for donor selection. Despite the potential for optimizing haploidentical outcomes through the use of the nomogram, the data indicated that matched unrelated donors consistently maximized predicted disease-free survival across all patient subgroups. This finding suggests that while the three-factor tool can refine the use of haploidentical grafts when necessary, a matched unrelated donor remains the preferred clinical choice for achieving the highest probability of survival without disease recurrence.

Identifying a Zone of Equivalence for Older Patients

The exploratory multivariable modeling further refined the clinical hierarchy between haploidentical donors and matched related donors, particularly regarding the influence of the HLA-B leader peptide. This peptide is a signal sequence that dictates the expression and presentation of HLA-B molecules, thereby influencing Natural Killer cell alloreactivity (the immune response triggered by differences between donor and recipient cells). The researchers found that in comparisons between haploidentical and matched related donors, B-leader mismatch appeared to favor matched related donors across all age groups. This suggests that when a genetic mismatch at the B-leader exists, the traditional matched related donor remains the superior choice for optimizing patient outcomes, regardless of whether the recipient is young or old. A distinct clinical scenario emerged when the analysis focused on the subset of older patients aged 60 years or older. In this specific demographic, B-leader matched haploidentical donors modeled near-equivalent disease-free survival to matched related donors. This finding identifies a specific intersection of advanced recipient age and precise immunogenetic matching where the historical disadvantage of haploidentical transplantation may be neutralized. The authors characterized this phenomenon as a hypothesis-generating zone of equivalence, noting that while the modeled outcomes are comparable, this zone of equivalence requires prospective validation through controlled clinical trials to confirm if B-leader matched haploidentical grafts can truly serve as a functional substitute for matched sibling donors in the elderly population.

Study Info

Uncoupling toxicity from efficacy in donor selection for allogeneic hematopoietic cell transplantation: a retrospective cohort study using double machine learning

Rohtesh S. Mehta, Yosra M. Aljawai, Warren Fingrut, Portia Smallbone, et al.

Journal Journal of Hematology & Oncology

Published April 20, 2026