Establishment of a machine learning-based prediction framework to assess trade-offs in decisions that affect post-HCT outcomes

In this study, we propose a conceptual framework of decision support tools, built upon machine learning and multi-objective optimization, aimed at offering a deeper understanding of the complex trade-offs involved in hematopoietic stem cell transplantation (HCT) across various blood diseases. Our main contribution is in proposing a means to assess benefits and risks across choices that affect multiple outcomes post-HCT, such as overall survival, event-free survival, rejection, relapse, and graft-versus-host disease. We elaborate on the development of machine learning models for predicting HCT outcomes, discuss the potential insights these models might offer, and propose how decision support tools could be informed by these insights. Our framework is demonstrated using an extensive Center for International Blood and Marrow Transplant Research (CIBMTR) HCT dataset.