Machine learning models using longitudinal laboratory data predict survival in allogeneic stem cell recipients

Allogeneic hematopoietic stem cell transplantation (HSCT) remains a potentially curative therapy for hematologic malignancies and bone marrow failure syndromes, yet outcomes remain highly heterogeneous and existing prognostic tools rely predominantly on pre-transplant clinical characteristics. We retrospectively analyzed 500 consecutive patients who underwent allogeneic HSCT at a single academic center between 2019 and 2025. Clinical characteristics, transplant variables, comorbidities, medication exposures, and longitudinal laboratory parameters were evaluated for associations with overall survival. Machine learning models incorporating dynamic post-transplant laboratory parameters demonstrated high performance in predicting overall survival compared to existing risk stratification techniques. Among evaluated predictors, early hematologic recovery emerged as one of the strongest determinants of long-term outcome. Specifically, achievement of a platelet count ≥90,000/μL by day +30 was independently associated with superior overall survival. These findings demonstrate that dynamic post-transplant laboratory data provide clinically meaningful prognostic information beyond traditional pre-transplant risk factors. Integration of early hematologic recovery metrics into machine learning-based prediction models may enable more accurate identification of high-risk patients and support individualized surveillance and management strategies following allogeneic HSCT.