Aurélien Latouche



Regression modeling of the cumulative incidence function with missing causes of failure


Abstract: Methods for estimating the effects of prognostic factors on the risk of death from a given cause often assume that the cause of death is known for all patients. Exclusion of individuals with missing cause of death information might lead to biased estimates. Some authors have proposed methods taking into account the missing cause of death mechanism, particularly for modeling the cause-specific hazards. However, little attention has been given to direct modeling of the cumulative incidence function which is of prime interest with competing risks. The rationale for the present work is to derive a flexible class of regression models for the cumulative incidence function when there are missing causes of failure. More precisely, we propose two approaches that extend the Andersen-Klein model to the missing cause setting. The first approach is grounded on the inverse probability weighting paradigm for dealing with missing data and the second is a multiple imputation method tailored for the Andersen-Klein model. We illustrate both approaches by analyzing the data from the ECOG 1178 breast cancer treatment clinical trial.

Updated on 12/03/2014


Access to all events