Abstract:
|
This paper presents a likelihood-based approach for studies with repeated measures and informative right censoring. The longitudinal events and censoring process share a bivariate latent random variable that corresponds to the random intercept and slope. Individual observations are modeled to allow inclusion of subjects with only one measurement. Censoring process and longitudinal observations are jointly modeled in a likelihood function that is integrated over the random effects. The log of the marginal likelihood function is maximized to obtain maximum likelihood estimation for the population slopes and intercepts, their variance-covariance matrix, and their respective censoring parameters, and empirical Bayes estimates for individual slopes and intercepts. Simulation study was conducted to assess the performance of the model using different underlying censoring distribution and considering various levels of censoring. Sensitivity analysis was also carried out to verify the robustness of the model for misspecification of censoring distribution and normality assumption. This model was applied on two clinical datasets to assess progression of kidney disease over time.
|