Title: Joint modelling of longitudinal data involving time-varying covariates
Authors: Reza Drikvandi - Manchester Metropolitan University (United Kingdom) [presenting]
Abstract: Longitudinal studies often produce data with both time-invariant and time-varying covariates. There are several major challenges with analysing such longitudinal data. For example, similar to classical regression models, standard models for longitudinal data ignore the covariate process and treat all covariates as fixed variables, while like the response variable, time-varying covariates also change over time and their evolution over time could provide important information. Also, longitudinal models assume that the response variable measured at each follow-up time depends on the time-varying covariates measured at that time point only, but the response variable could also depend on the previous measurements of time-varying covariates. We introduce a novel joint mixed model for the longitudinal outcome and the time-varying covariates to overcome such challenges. We use random effects to account for the association between the response and the time-varying covariates. We also incorporate P-spline functions into the joint model to capture the evolutions of the longitudinal response and the time-varying covariates over time. The proposed method is investigated theoretically and practically, and motivated by data from an AIDS cohort study in which HIV+ patients have CD4 cell count and viral load measured at repeated visits before and after receiving treatment.