Flexible longitudinal linear mixed models for multiple censored responses data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Biomedical studies often involve repeated measures with detection limits, leading to censored data.
• Analyzing multiple series of longitudinal data with heavy tails and irregular intervals presents statistical challenges.
• Existing multivariate censored linear mixed effect (MLMEC) models may not adequately handle these complexities.

Purpose of the Study:

• To develop a robust generalization of the MLMEC model for analyzing multivariate, irregularly observed, censored longitudinal data.
• To incorporate heavy-tailed distributions and autocorrelation structures into the modeling framework.
• To propose an exact estimation procedure for parameter estimation and model diagnostics.

Main Methods:

• Developed a robust MLMEC model using scale mixtures of normal distributions.
• Incorporated a damped exponential correlation structure for irregularly observed measures.
• Employed a stochastic approximation of the Expectation-Maximization (EM) algorithm for maximum-likelihood estimation.

Main Results:

• The proposed method provides an exact estimation procedure for fixed effects and variance components.
• The approach facilitates quick parameter estimation and calculation of standard errors.
• The method yields predictions of unobservable response values and the log-likelihood function as byproducts.

Conclusions:

• The robust MLMEC model effectively analyzes complex longitudinal data with censored and heavy-tailed distributions.
• The proposed estimation procedure is efficient and provides valuable statistical outputs.
• The method demonstrates utility in analyzing real-world biomedical data, such as AIDS clinical trials.