Competing risks analysis of time-to-event data for cardiovascular surgeons

Abstract

ObjectivesThe purpose of this article is to provide thoracic and cardiovascular surgeons with a practical step-by-step strategy to use in collaboration with a biostatistician for implementation of competing risks analysis when analyzing time-to-event data. Patients may have an outside event that precludes the event of interest. Traditional time-to-event analysis incorrectly assumes noninformative censoring in this scenario, which will lead to invalid results and conclusions. MethodsThe steps are (1) to determine whether competing risks analysis is needed, (2) to perform a nonparametric analysis, (3) to perform a model-based analysis, (4) to interpret the results, and (5) to compare to traditional survival analysis methods. We apply our approach to a hypothetical cardiovascular surgery example in determining the hazard of mortality after the stage 3 Fontan operation associated with prematurity among patients with hypoplastic left heart syndrome who had successful completion of Norwood stage 1 while incorporating mortality during the stage 2 bidirectional Glenn procedure as a competing risk. We apply nonparametric, semiparametric, and parametric methods. ResultsAlthough Cox regression establishes prematurity as a significant risk factor of mortality after stage 3 (hazard ratio, 1.26; 95% confidence interval, 1.06-1.50; P = .009), the competing risks analysis with the Fine-Gray model accounting for mortality after stage 2 determines that prematurity is not a significant predictor (hazard ratio, 1.07; 95% confidence interval, 0.90-1.27; P = .467). ConclusionsThis article provides a practical step-by-step approach for making competing risks more accessible for cardiac surgeons collaborating with a biostatistician in analyzing and interpreting time-to-event data.