Fig. 1. Study Flowchart.

Four hundred patients were randomly selected from the Paris Transplant Group database, a multicenter prospective cohort involving 4000 patients consecutively transplanted between 2005 and 2014 from four French academic transplant centers (Necker and Saint-Louis hospitals, Paris; Foch hospital, Suresnes; Toulouse hospital, Toulouse). Forty-four parameters from the first year of transplantation were included and integrated into an anonymized Electronic Health Record (EHR) for evaluation by nine transplant physicians and the iBox prediction system. Allograft survival at seven years post-evaluation predicted by the physicians and the iBox were compared to the observed allograft survival to assess prediction performances. The agreement of physicians’ predictions was compared using intraclass correlation, and the most important parameters in physicians’ prediction were ranked and compared using mean decreased accuracy from a random forest algorithm and Fleiss kappa. EHR: electronic health record.