Bayesian weighted composite quantile regression in varying-coefficient partially linear mixed-effects model with ordinal longitudinal data

In real applied fields such as clinical medicine, epidemiology and biology, we often encounter longitudinal data with ordinal responses. It is rather common to observe ordinal longitudinal data that simultaneously exhibit multiple characteristics, including individual heterogeneity, nonlinear trends, time-dependent effects and skewed distributions. Modeling such ordinal longitudinal data with all the above characteristics and conducting statistical inference analysis constitute a difficult problem. In order to better interpret the individual heterogeneity, depict the dynamic change characteristics and skewed distributions in ordinal longitudinal response and obtain more reliable and robust inferential results in practical applications, we attempt to propose a Bayesian weighted composite quantile regression varying-coefficient partially linear mixed-effects model, which adopts the ordinal latent variable inference framework to deal with ordinal responses and uses B-splines to approximate the varying-coefficient functions. A joint Bayesian hierarchical model for weighted composite quantile regression is established and an efficient Gibbs sampling algorithm is developed for posterior inference by introducing the mixture representation of the composite asymmetric Laplace distribution. Finally, Monte Carlo simulation studies are used to illustrate the robustness and flexibility of the proposed method, and it is applied to an ordinal longitudinal data set of prostate cancer.