Competing risk problems with no independence assumed: Does it make a difference?

Abstract

In survival analysis the Kaplan Meier estimator (Kaplan and Meier, 1958) is widely used to provide information about the covariates effect on a response. Whenever a competing risks framework is given, the Kaplan Meier estimator may overstimate or underestimate the effect of the covariates (Di Serio, 1997) due to the unrealistic underlying independence assumptions. The cumulative incidence curve (CIC) is suggested as alternative estimator in these frameworks. Results deriving from the use of the CIC applied to esophageal cancer data show how this alternative tool is an intuitive as the Kaplan Meier estimator and does not required any further statistical assumptions for identifiability.