Differential impact of CD34+ cell dose for different age groups in allogeneic hematopoietic cell transplantation for acute leukemia: a machine learning-based discovery.

Allogeneic hematopoietic cell transplantation (allo-HCT) presents a potentially curative treatment for hematologic malignancies yet carries associated risks and complications. Continuous research focuses on predicting outcomes and identifying risk factors. Notably, the influence of CD34+ cell dose on overall survival (OS) has been the subject of numerous studies yielding contradictory results. We developed machine learning (ML) models to predict allo-HCT outcomes and, through the application of SHapley Additive exPlanations (SHAP), an explainable artificial intelligence (XAI) technique enabled the identification of new and clinically relevant feature-outcome relationships. In particular, we identified a clear interaction between CD34+ cell dose of peripheral blood stem cell (PBSC) grafts and patient age at allo-HCT for patients with acute leukemia. Results of multivariable analysis validated the interaction effect: in young patients with acute leukemia (aged ≤45 years), low dose of CD34+ cells (<4.3 × 10 CD34+/kg) was associated with better OS against high dose (≥7 ×10 CD34+/kg) (hazard ratio [HR], 0.38; p = 0.019), while for older patients with acute leukemia (>45 years), low CD34+ cell dose (<3.8 ×10 CD34+/kg) was associated with worse OS against high dose (≥6.1 ×10 CD34+/kg) (HR, 1.58; p = 0.033). In conclusion, our findings suggest that tailoring CD34+ cell dose by patient age may benefit patients with acute leukemia undergoing allo-HCT, while XAI showcases excellent proficiency in revealing such interactions.