PREDICTION MODELING OF KIDNEY TRANSPLANT OUTCOME by

Although renal transplant is the preferred modality for end-stage renal disease, it brings with it a number of challenges primarily associated with lack of individualized approach. The goals of the present project were: (1) to determine the most significant and clinically practical predictors of kidney transplant outcomes (patient survival, allograft survival, posttransplant complications) using United States Renal Data System (USRDS) data; (2) based on the selected predictors, to generate prediction models of renal transplant outcomes. Our initial study developed prediction models using logistic regression and treebased algorithms derived from data provided by the United Network of Organ Sharing (UNOS). A series of follow-up projects, using data supplied by the United States Renal Data System (USRDS), was performed. We were able to capture significant associations between donor, recipient, and transplant procedure variables (that could not be derived from UNOS data) and the allograph and recipient survival. Among our important findings, compared to peritoneal dialysis (PD), hemodialysis is associated with increased risk of graft failure and recipient death; preemptive retransplantation is associated with an increased risk of graft failure; increased time on dialysis between transplants is associated with a negative effect upon graft and recipient survival in most patient subgroups; short-term (6 months or less) dialysis had no negative effect on graft survival compared to preemptive transplants; certain socioeconomic factors, such as higher education level, citizenship, and type of insurance coverage, influenced graft and recipient outcomes, independent of racial differences; and that one particular