Validation in a single-center cohort of existing predictive models for delayed graft function after kidney transplantation

Background: Kidney transplantation is the preferred treatment for patients with end-stage renal disease. Delayed graft function (DGF) is a common complication and is associated with short-and long-term outcomes. Several predictive models for DGF have been developed. Material/Methods: 497 kidney transplantations from deceased donors at our center between 2005-2011 are included. Firstly, the predictive accuracy of the existing models proposed by Irish et al. (M1), Jeldres et al. (M2), Chapal et al. (M3), and Zaza et al. (M4) was assessed. Secondly, the existing models were aggregated into a meta-model (MM) using stacked regressions. Finally, the association between 47 risk factors and DGF was studied in our - cohort-fitted model (CFM) using logistic regression. The accuracy of all models was assessed by area under the receiver operating characteristic curve (AUROC) and Hosmer-Lemeshow test. Results: M1, M2, M3, M4, MM, and CFM have AUROCs of 0.78, 0.65, 0.59, 0.67, 0.78, and 0.82, respectively. M1 (P=0.018), M2 (P<0.001), M3 (P<0.001), and M4 (P<0.001) overestimate the risk. MM (P=0.255) and CFM (P=0.836) are well calibrated. Donor subtype (P<0.001), recipient cardiac function (P<0.001), donor serum creatinine (P