Improving Deceased Donor Kidney Utilization: Predicting Risk of Nonuse with Interpretable Models

Abstract

Background. Despite the increasing disparity between the number of patients awaiting kidney transplants and the availability of deceased donor kidneys, a significant number of donated kidneys go unused. Early identification of organs at high risk of nonuse can facilitate effective allocation interventions, ensuring these organs are offered to patients who could potentially benefit from them. While several machine learning models have been developed to predict nonuse risk, the complexity of these models compromises their practical implementation. Methods. We propose implementable nonuse risk prediction models that consist of a minimal set of variables, including the Kidney Donor Risk Index (KDRI), along with factors selected by machine learning models or transplantation experts. Our approach takes into account the influence of Organ Procurement Organization (OPO) behavior on kidney disposition. Results. The proposed models demonstrate competitive performance compared to more complex models that involve a large number of variables. Importantly, they maintain simplicity and interpretability. Conclusions. Our results provide accurate risk predictions, offer valuable insights into key factors contributing to kidney nonuse, and underscore significant variations among OPOs in the allocation of hard-to-place kidneys. These findings can inform the design of effective organ allocation interventions, increasing the likelihood of transplantation for hard-to-place kidneys.