New SRTR Risk-Adjustment Models for Adult Kidney Transplant Outcomes

J. Snyder,¹ N. Salkowski,¹ D. Zaun,¹ H. Xiong,¹ S. Leppke,¹ B. Thompson,¹ D. Segev,^1,2 L. Hunsicker,¹ A. Israni,¹ B. Kasiske.¹

¹Scientific Registry of Transplant Recipients, Minneapolis Medical Research Foundation, Minneapolis, MN
²Johns Hopkins University, Baltimore, MD.

Meeting: 2015 American Transplant Congress

Abstract number: 62

Keywords: Kidney transplantation, Risk factors

Topic:

Following a recommendation of the Consensus Conference on Transplant Program Quality and Surveillance, SRTR implemented a 3-year rolling cycle to rebuild the statistical models used to derive expected outcomes and evaluate program performance. SRTR developed a standardized model-building process with oversight by the SRTR Technical Advisory Committee, and new models for kidney transplant recipients have been developed. Transplants performed 1/1/2011-6/30/2013 were used to develop the models. The least absolute shrinkage and selection operator (LASSO) procedure was used to determine the optimal set and form of risk predictors to include in the models such that the predictive ability of the models was optimized using 10-fold cross-validation. The discriminatory ability of the models as measured by the C statistic ranged from 0.65 to 0.76 (Table), comparable with the previous models.

Model Discriminatory Ability (C statistics)

Model	C statistic
Adult, deceased donor, graft survival	0.6576
Adult, deceased donor, patient survival	0.7136
Adult, living donor, graft survival	0.6735
Adult, living donor, patient survival	0.7599

The new models include an increased number of risk predictors. For example, the previous deceased donor graft survival model included 13 donor factors and 10 recipient factors. The new model includes 21 donor factors and 14 recipient factors. Donor factors include kidney donor risk index (KDRI), demographic characteristics, laboratory values, viral serologies, donor management characteristics, and HLA match. Recipient characteristics include demographic characteristics, laboratory values, comorbid conditions, viral serologies, procedure type, and CPRA. The calibration of the new models was good such that expected numbers of graft failures coincided well with observed numbers of graft failures when recipients were categorized into 20 groups with similarly predicted risk (Figure). Exploration of additional risk factors may further improve risk prediction in the future.

« Back to 2015 American Transplant Congress