A general joint latent class model of longitudinal and survival data with the covariance modelling

Abstract: Based on the proposed time-varying JLCM (Miao and Charalambous, 2022), the heterogeneous random covariance matrix can also be considered, and a regression submodel for the variance-covariance matrix of the multivariate latent random effects can be added to the joint latent class model. A general joint latent class model with heterogeneous random-effects modelling is a natural extension of the time-varying JLCM, which consists of the linear and the log link functions to model the covariance matrices as the variance-covariance regression submodel based on the modified Cholesky decomposition, longitudinal submodel, survival submodel as well as the membership probability. It can help to get more information from the random covariance matrix through the regression submodel and get unbiased estimates for all parameters by modelling the variance-covariance matrix. By adding the regression model, the homogeneous random effects assumption can be tested and the issue of high-dimensional heterogeneous random effects can be easily solved. The Bayesian approach will be used to estimate the data. DIC value is the criterion for deciding the optimal k value. We illustrate our general JLCM on a real data set of AIDS study and we are interested in the prospective accuracy of our proposed JLCM as well as doing the dynamic predictions for time-to-death in the joint model using the longitudinal CD4 cell count measurements.