Re-assessing prolonged cold ischemia time in kidney transplantation through machine learning consensus clustering.

ABSTRACT

BACKGROUND: We aimed to cluster deceased donor kidney transplant recipients with prolonged cold ischemia time (CIT) using an unsupervised machine learning approach. METHODS: We performed consensus cluster analysis on 11 615 deceased donor kidney transplant patients with CIT exceeding 24 h using OPTN/UNOS data from 2015 to 2019. Cluster characteristics of clinical significance were identified, and post-transplant outcomes were compared. RESULTS: Consensus cluster analysis identified two clinically distinct clusters. Cluster 1 was characterized by young, non-diabetic patients who received kidney transplants from young, non-hypertensive, non-ECD deceased donors with lower KDPI scores. In contrast, the patients in cluster 2 were older and more likely to have diabetes. Cluster 2 recipients were more likely to receive transplants from older donors with a higher KDPI. There was lower use of machine perfusion in Cluster 1 and incrementally longer CIT in Cluster 2. Cluster 2 had a higher incidence of delayed graft function (42% vs. 29%), and lower 1-year patient (95% vs. 98%) and death-censored (95% vs. 97%) graft survival compared to Cluster 1. CONCLUSIONS: Unsupervised machine learning characterized deceased donor kidney transplant recipients with prolonged CIT into two clusters with differing outcomes. Although Cluster 1 had more favorable recipient and donor characteristics and better survival, the outcomes observed in Cluster 2 were also satisfactory. Overall, both clusters demonstrated good survival suggesting opportunities for transplant centers to incrementally increase CIT.