Haploidentical versus matched unrelated donor transplantation with post-transplant cyclophosphamide: a platform-dependent machine learning analysis of donor age.

In allogeneic hematopoietic cell transplantation with post-transplant cyclophosphamide (PTCy), clinicians frequently face a critical choice between a readily available, often younger, haploidentical and a fully matched unrelated donor (MUD). The platform-specific influence of donor age on survival is a critical, unquantified factor that complicates clinical decision-making. We retrospectively analyzed 4258 adult patients with acute leukemia who underwent first allogeneic HCT with PTCy (2017-2021). We employed machine learning (Random Survival Forests and DeepSurv) alongside robust regression models, including Inverse Probability of Treatment Weighting, 1:1 Propensity Score Matching, and Elastic-Net penalized Cox regression. Machine learning models revealed a divergent association of donor age with survival depending on donor type. The MUD-PTCy platform proved remarkably resilient; a 1% absolute increase in mortality risk (equivalent to Number-Needed-to-Harm of 100) did not emerge in donors up to age 50. In stark contrast, the age-sensitive Haploidentical cohort reached this same risk threshold at a donor age of just 38 years. MUD-PTCy was independently associated with a significant overall survival advantage (Adjusted Hazard Ratio 0.85; 95% confidence interval, 0.75-0.97; P = 0.01). This analysis provides a quantitative framework to guide the trade-off between HLA matching and donor age, supporting individualized decision-making and a rationale to reconsider restrictive donor age policies.