Competing and Noncompeting Risk Models for Predicting Kidney Allograft Failure

Area of Science:

• Nephrology
• Biostatistics
• Clinical Epidemiology

Background:

• Prognostic models are vital for clinical trials and patient management.
• The impact of competing risks on model performance is under-investigated.
• Kidney transplantation involves competing risks: allograft failure and death with a functioning graft.

Purpose of the Study:

• To assess the performance of competing risk and non-competing risk models in kidney transplantation.
• To compare model development, stability, discrimination, calibration, fit, utility, and generalizability.
• To provide an exhaustive assessment using over 15 metrics.

Main Methods:

• Included 11,046 kidney transplant recipients from 10 countries.
• Developed prediction models for graft failure using Cox, Fine-Gray, and cause-specific Cox regression.
• Assessed models using a transparent framework for competing and non-competing risk modeling.

Main Results:

• 1,497 (14%) graft failures and 1,003 (9%) deaths occurred.
• Both model types showed similar and stable risk estimates for graft failure.
• Discrimination and overall fit were comparable, with concordance indices from 0.76 to 0.87.

Conclusions:

• Competing and non-competing risk models demonstrated similar performance in predicting long-term kidney graft failure.
• Models performed well across various subpopulations and clinical scenarios.
• A competing risk approach showed a trend towards improved calibration in some high-risk groups.