Bayesian multivariate nonlinear mixed models for censored longitudinal trajectories with non-monotone missing values

• Published in: Metrika Vol. 87; no. 5; pp. 585 - 605
• Main Authors: Wang, Wan-Lun, Castro, Luis M., Lin, Tsung-I
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2024; Springer Nature B.V
• Subjects: Algorithms; Bayesian analysis; Economic Theory/Quantitative Economics/Mathematical Methods; Markov chains; Mathematics and Statistics; Multivariate analysis; Probability Theory and Stochastic Processes; Statistics; Missing data imputation; Censored data recovery; Posterior sampling; Truncated multivariate normal distribution; Markov chain Monte Carlo
• ISSN: 0026-1335; 1435-926X
• DOI: 10.1007/s00184-023-00929-x

The analysis of multivariate longitudinal data may often encounter a difficult task, particularly in the presence of censored measurements induced by detection limits and intermittently missing values arising when subjects do not respond to a part of outcomes during scheduled visits. The multivariate nonlinear mixed model (MNLMM) has emerged as a promising analytical tool for multi-outcome longitudinal data following arbitrarily nonlinear profiles with random phenomena. This article presents a generalization of the MNLMM, called MNLMM-CM, designed to simultaneously accommodate the effects of censorship and missingness within a Bayesian framework. Specifically, we develop a Markov chain Monte Carlo procedure that combines a Gibbs sampler with the Metropolis–Hastings algorithm. This hybrid approach facilitates Bayesian estimation of essential model parameters and imputation of non-responses under the missing at random mechanism. The issue of posterior predictive inference for the censored and missing outcomes is also addressed. The effectiveness and performance of the proposed methodology are demonstrated through the analysis of simulated data and a real example from an AIDS clinical study.