Figure 1

Study design. Data were obtained from the United Network for Organ Sharing/Organ Procurement and Transplantation Network (UNOS/OPTN) on adult deceased donor kidney transplant recipients transplanted between January 1, 2007, and May 31, 2012, for a development set and between June 1, 2012, and December 31, 2012, for a validation set. First, recursive feature elimination with random forest was applied to the data pre-processed as shown. The cleaned dataset was under-sampled to adjust the class distribution, resulting in a dataset of n = 25,604. Five widely-used machine learning algorithms were then trained on the data using tenfold cross-validation. Finally, each model was assessed in the validation cohort for discrimination, calibration, and clinical utility. The letters n and m represent the numbers of records and features respectively. IQR interquartile range.