Pre-transplant and transplant parameters predict long-term survival after hematopoietic cell transplantation using machine learning

Background Allogeneic hematopoietic stem transplantation (allo-HSCT) constitutes a curative treatment for various hematological malignancies. However, various complications limit the therapeutic efficacy of this approach, increasing the morbidity and decreasing the overall survival of allo-HSCT recipients. In everyday clinical practice, various laboratory and clinical biomarkers and scorning systems have been developed and implemented focusing on the recognition of high-risk patients for organ dysfunction-related complications and those who might experience low overall survival. However, the predictive accuracy of developed scores has been reported deficient in some studies. The aim of the current retrospective study is to develop a machine learning (ML) model to predict the long-term survivorship of patients who receive allo-HSCT based on clinical pre- and post-allo-HSCT variables, and on transplantation-related characteristics. Methods For this purpose, a database of 564 allo-HSCT recipients incorporating 16 clinical and laboratory variables and the survivorship status of the patients during follow-up (Alive, Dead, Alive but follow-up less than 24 months) was used. An ML model was developed and tested, based on the previously published Data Ensemble Refinement Greedy Algorithm (DEGRA) algorithm. Results A predictive ML model was built with 92.02 % accuracy. The eight parameters included in the algorithm were the following: CD34+ cells infused, patients' age and gender, conditioning regimen toxicity, disease risk index (DRI), graft source, and platelet and neutrophil engraftment. Conclusion To our knowledge, this is the first AI model incorporating post-HSCT variables for the prediction of mortality in adult HSCT recipients. In the era of precision medicine, the recognition of patients who undergo allo-HSCT and face a great risk for mortality and morbidity, with high-accuracy algorithms is crucial.