Federated Transfer Learning for a Centralized Organ Donation Repository

Insufficient organ availability for donation fails to meet the increasing demand for transplants, leading to thousands of deaths annually among those awaiting organs. One organ donor has the potential to save eight lives and positively influence more than 75 others. While 50% of children over 1 year old, who cease life-sustaining therapy, can donate organs after cardiac death, addressing the gap between organ demand and supply necessitates meticulous patient/donor selection and matching. In our paper, we study how machine learning tools like the Cox Proportional Hazard model, Random Forest, and XGBoost can be used in a centralized organ donation database. We also examine the effectiveness of Federated Learning and survival analysis as tools to enhance predictive modelling for organ donation outcomes. Our research indicates that Federated Transfer Learning surpasses other machine learning techniques in terms of accuracy. These findings propose that integrating federated transfer learning into a centralized organ donation repository can significantly enhance the system's predictive capabilities, offering a promising approach for advancing organ donation management and allocation. This information is preserved for future use in predicting potential issues. The developed federated transfer learning model demonstrates commendable accuracy in organ donation, with a value of 0.998. Consequently, implementing such a model can aid hospitals in their mission to save lives.