External Validation of Clinical Prediction Models of Patient and Graft Survival in a Canadian Cohort of Kidney Transplant Recipients
1Division of Nephrology, Department of Medicine, University of Toronto, Toronto, ON, Canada
2Institute of Health Policy, Management and Evaluation, University of Toronto, Toronto, ON, Canada
3Division of Nephrology, Sunnybrook Health Sciences Center, Toronto, ON, Canada
4Division of Nephrology and the Kidney Transplant Program, University Health Network, Toronto, ON, Canada
5Division of Nephrology and the Renal Transplant Program, St. Michael's Hospital, Toronto, ON, Canada.
Meeting: 2015 American Transplant Congress
Abstract number: 444
Keywords: Graft failure, Kidney transplantation, Prediction models, Survival
Session Information
Session Name: Concurrent Session: Kidney - Delayed Graft Function and Older Age
Session Type: Concurrent Session
Date: Tuesday, May 5, 2015
Session Time: 4:00pm-5:30pm
Presentation Time: 4:48pm-5:00pm
Location: Room 115-AB
Background: Several clinical prediction models (CPM) of patient and graft survival in kidney transplant recipients have been developed as tools to assist physicians in identifying high-risk patients. However, the performance of these models is often not tested in external populations. The purpose of this study was to validate selected CPM in a large cohort of Canadian kidney transplant recipients.
Methods: Our external validation cohort consisted of 1,326 adult kidney transplant recipients alive with a functioning graft at 1-year post transplant from the Comprehensive Renal Transplant Research Information System (CoReTRIS) database, which includes all kidney transplant recipients from 2000 to 2014 from Toronto General Hospital, Toronto, Canada. After systematically reviewing the literature for CPM of patient and graft survival, we selected three models (death with graft function (DWGF), total graft failure (TGF) and death-censored graft failure (DCGF)) with the best performance metrics, least risk of bias and model variables included in our dataset for external validation (Hernandez et al. Transplantation 2009; 88: 803-809 & Moore et al. AJKD 2011; 57(5): 744-751). Discrimination and calibration were assessed.
Results: Discrimination of all CPM was modest with a C-statistic of 0.69 for DWGF, 0.67 for TGF and 0.74 for DCGF. Model calibration was determined by comparing observed vs. predicted probabilities of the event of interest, which were discordant for all three models (Hosmer-Lemeshow statistic P < 0.001).
Conclusions: This study highlights the importance of validating CPM in external populations. Poorly performing models should be re-calibrated prior to clinical use, or alternatively, new prediction models should be developed when existing CPM perform poorly in external validations.
To cite this abstract in AMA style:
Singh S, Naimark D, Victor J, Kim1 J. External Validation of Clinical Prediction Models of Patient and Graft Survival in a Canadian Cohort of Kidney Transplant Recipients [abstract]. Am J Transplant. 2015; 15 (suppl 3). https://atcmeetingabstracts.com/abstract/external-validation-of-clinical-prediction-models-of-patient-and-graft-survival-in-a-canadian-cohort-of-kidney-transplant-recipients/. Accessed May 31, 2025.« Back to 2015 American Transplant Congress