Comparing Cumulative Incidence Functions Between Nonrandomized Groups in the Presence of Competing Risks

• Activity Number: 38
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Ludi Fan*+ and Douglas E. Schaubel
• Companies: University of Michigan and University of Michigan
• Keywords: Cumulative incidence function ; Competing risks ; Cox models ; Center effect

Abstract

In the presence of competing risks, it is often of interest to compare cumulative incidence functions (CIFs) among groups. If group is not randomized, meaningful group-specific contrasts require correction for imbalances in group-specific covariate distributions. A motivating example is evaluating organ procurement organizations (OPOs) with respect to the probability that a patient wait-listed for kidney transplantation receives a transplant, with death as a competing risk. We propose two methods for comparing group-specific CIFs to an overall average. Both methods involve comparing the CIF of each OPO to that of a hypothetical average OPO, giving an observed minus expected measure. The CIFs are estimated by averaging over fitted values from stratified Cox models of the cause-specific hazards. The second method replaces the observed term from the first method with a weighted count, so that results of the Cox models will only be used in aggregate form in the expected term. Large-sample properties are derived using martingale and empirical process theory, with finite-sample properties evaluated through simulation.We apply the methods to data from a national organ transplant registry.