The impact of competing risks in kidney allograft failure prediction

Abstract

ABSTRACTBackgroundPrognostic models are becoming increasingly relevant in clinical trials as potential surrogate endpoints, and for patient management as clinical decision support tools. However, the impact of competing risks on model performance remains poorly investigated. We aimed to carefully assess the performance of competing risks and non-competing risks models in the context of kidney transplantation, where allograft failure and death with a functioning graft are two competing outcomes.MethodsWe included 10 546 adult kidney transplant recipients enrolled in 10 countries (3941 patients in the derivation cohort, 6605 patients in international external validation cohorts). We developed prediction models for long-term kidney graft failure prediction, without accounting (i.e., censoring) and accounting for the competing risk of death with a functioning graft, using Cox and Fine-Gray regression models. To this aim, we followed a detailed and transparent analytical framework for competing and non-competing risks modelling, and carefully assessed the models’ development, stability, discrimination, calibration, overall fit, and generalizability in external validation cohorts and subpopulations. In total, 15 metrics were used to provide an exhaustive assessment of model performance.ResultsAmong the 3941 recipients included in the derivation cohort, 538 (13.65%) lost their graft and 414 (10.50%) died after a median follow-up post-risk evaluation of 5.77 years (IQR 3.52-7.00). In the external validation cohorts, 896 (13.56%) graft losses and 525 (7.95%) deaths occurred after a median follow-up post-risk evaluation of 4.25 years (IQR 2.35-6.59). At 7 years post-risk evaluation, overestimation of the cumulative incidence was moderate when using Kaplan-Meier, compared to the Aalen-Johansen estimate (16.71% versus 15.67% in the derivation cohort). Cox and Fine-Gray models for predicting the long-term graft failure exhibited similar and stable risk estimates (average MAPE of 0.0140 and 0.0138 for Cox and Fine-Gray models, respectively). At 7 years post-risk evaluation, discrimination and overall fit were good and comparable in the external validation cohorts (concordance index ranging from 0.76 to 0.86, Brier Scores ranging from 0.102 to 0.141). In a large series of subpopulations and clinical scenarios, both models performed well and similarly.ConclusionsCompeting and non-competing risks models performed similarly in predicting long-term kidney graft failure. These results should be interpreted in light of the low rate of the competing event in our cohort, and do not stand as a general conclusion for competing risks modelling. Depending on the clinical scenario and the population considered, competing risks may be crucial to consider for accurate risk predictions.