Uncoupling toxicity from efficacy in donor selection for allogeneic hematopoietic cell transplantation: a retrospective cohort study using double machine learning
Clinical Snapshot
PICO Framework
| P — Population | Adult patients undergoing allogeneic hematopoietic cell transplantation (n=1,713 discovery cohort, n=6,829 registry replication cohorts) |
| I — Intervention | Haploidentical donor transplantation |
| C — Comparator | Matched unrelated donor (MUD) and matched related donor (MRD) transplantation |
| O — Outcomes | Non-relapse mortality (NRM), relapse, disease-free survival, overall survival |
Bottom Line
This large retrospective cohort study using advanced machine learning techniques provides compelling evidence for personalized donor selection in allogeneic HCT. The key finding that cardiovascular/cerebrovascular comorbidities selectively amplify haploidentical donor toxicity (doubling mortality risk) while having minimal impact on MUD/MRD outcomes represents a clinically actionable insight. The proposed 3-factor nomogram identifying an 'Optimized Haploidentical' phenotype could guide donor selection, particularly for older patients (≥60 years) with B-leader matched haploidentical donors showing near-equivalent outcomes to matched related donors. While MUD donors appeared optimal across all subgroups, the identification of specific patient populations where haploidentical transplantation performs comparably offers important clinical utility given donor availability constraints. The biological uncoupling of relapse (disease-driven) from NRM (physiology-driven) provides a rational framework for individualized treatment decisions. However, prospective validation of these machine learning-derived recommendations is essential before widespread clinical implementation.
Key Findings
P Value: Not reported in abstract
Effect Size: Vascular comorbidities doubled haploidentical mortality risk (sub-distribution hazard ratio 2.09)
Primary Outcome: Non-relapse mortality and relapse as competing risks
Nnt Or Sensitivity: 3-factor nomogram identifies 'Optimized Haploidentical' phenotype with near-equivalent DFS to MRD in patients ≥60 years with B-leader matching
Confidence Interval: 95% CI 1.26-3.35
Clinical Application
Nomogram appears clinically feasible using readily available variables (age, vascular comorbidity, B-leader status) Highly relevant to Australian HCT programs; may inform donor selection algorithms and could influence PBS funding decisions for HLA typing and donor search strategies Adult patients requiring allogeneic HCT with available donor options
Abstract
Haploidentical hematopoietic cell transplantation (HCT) offers potent graft-versus-tumor effects but is historically limited by higher non-relapse mortality (NRM) compared to matched unrelated (MUD) or related donors (MRD). To algorithmically decouple the competing risks of NRM and relapse, we analyzed 1,713 adult patients (MD Anderson discovery cohort) and 6,829 patients (registry replication cohorts) using a double machine-learning framework of Causal Survival Forests to estimate individualized treatment effects, while Random Survival Forests defined absolute prognostic risk. We found biological uncoupling of relapse and NRM: relapse was driven by disease biology, while NRM was governed by physiological reserve. Specifically, cardiovascular or cerebrovascular comorbidities acted as selective amplifiers of haploidentical toxicity-a "vascular penalty" that doubled mortality risk in the bootstrapped model (sub-distribution hazard ratio 2.09, 95% confidence interval 1.26-3.35). We developed a 3-factor nomogram (recipient age, vascular comorbidity, B-leader matching status) to identify an "Optimized Haploidentical" phenotype. Exploratory multivariable modeling suggested MUD consistently maximized predicted disease-free survival across all subgroups. Among haploidentical versus MRD comparisons, B-leader mismatch appeared to favor MRD across ages, while B-leader matched haploidentical donors in older patients (≥ 60 years) modeled near-equivalent DFS to MRD, representing a hypothesis-generating "zone of equivalence" requiring prospective validation.
Related Research
Human vaccines & immunotherapeutics
Artificial intelligence in immune checkpoint inhibitor research: A bibliometric analysis of the landscape
30 July 2026
Blood advances
Successful gene therapy for transfusion-dependent α-thalassemia: a case report
29 July 2026
Journal of chemical information and modeling
Functionally Guided Graph Learning for Robust Cross-Patient Cell-Type Annotation in Single-Cell RNA Sequencing
29 July 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service