A0326
Title: Advancements in efficient estimation for mixed effects models with censored data
Authors: Shakhawat Hossain - University of Winnipeg (Canada) [presenting]
Abstract: Longitudinal and repeated measures data are frequently analyzed using mixed models. However, the presence of censored responses, a common occurrence in biomedical research and clinical trials due to detection limits, introduces complexity. These limits can result in left- or right-censored measurements. While adapted linear mixed effects models are often employed to handle such data, a likelihood-based approach is proposed for fitting a linear mixed effects model with normally distributed errors. An expectation-maximization algorithm is utilized for unrestricted maximum likelihood estimation. Furthermore, scenarios where model parameters are subject to uncertain linear constraints are explored, leading to the development of a restricted estimator. To improve the estimation of fixed effects, two refined estimator sets are introduced: A pretest estimator, and shrinkage and positive shrinkage estimators. The performance of these proposed estimators is evaluated against the unrestricted maximum likelihood estimator via extensive simulations and an application to longitudinal data from the AIDS Clinical Trials Group protocol A5055 study.
