Slope Estimation for Informatively Right-Censored Longitudinal Data Assuming a Nonconstant Survival Model for the Censoring Process: Application to Renal Transplant Patients

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Miran Jaffa*+ and Ayad A. Jaffa
• Companies: American University of Beirut and American University of Beirut
• Keywords: discrete survival model ; informative right censoring ; longitudinal data ; random effects ; slope estimation

Abstract

Informative right censoring is rampant in longitudinal studies and should be accounted for in the analysis to attain unbiased slope estimates. Here we propose a novel approach that jointly models the slopes and the censoring process in a likelihood function that is integrated over the random effects. Maximization of the marginal likelihood function generates maximum likelihood estimates for the population slope and empirical Bayes estimates for the individual slopes that are predicted using Gaussian quadrature. A survival model with logistic non-constant hazard in which the dropout probability is changing over time was adopted for the censoring process. Our simulation results indicated that this model decreased the associated bias and mean squared errors by about 50% compared to two-part, bootstrapping and mixed models. Our proposed approach was applied to renal transplant patients using the estimated glomerular filtration rate as the outcome of interest. These patients are prone to graft rejection that leads to informative right censoring. Population and individual slopes were estimated and high risk patients for graft rejection were identified based on their individual slopes.