CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Weighted average approach for joint modelling in absence of knowledge regarding the association structure Authors:  Maha Alsefri - University of Liverpool (United Kingdom) [presenting]
Ruwanthi Kolamunnage-Dona - University of Liverpool (United Kingdom)
Maria Sudell - University of Liverpool (United Kingdom)
Abstract: Over the last decade, there has been an increasing interest in applying joint models to related longitudinal and time-to-event outcome data in clinical research, especially in studies with interest in examining patients repeatedly until an event of interest. The usage of this method is increasing due to their ability to account for informative dropout in the longitudinal data and to allow for the inclusion of time-varying covariates measured with error in the survival model. However, according to recent reviews of joint modelling, the current research is considerably limited in terms of specifying the association structure between the longitudinal and time-to-event outcomes in the absence of background knowledge. Information criteria (such as DIC) are generally applied to identify the best fit joint model from several potential association structures. We propose an alternative weighted averaging approach, which can be utilized to combine estimations from potential joint models. This results in parameter estimation being based on multiple different association structures instead of limiting to just one, possibly incorrect. Simulated data is used to investigate the proposed approach in both frequentist and Bayesian settings. Results of the simulation study, and real-world application from PBC (primary biliary cirrhosis) and HCC (Hepatocellular Carcinoma) studies, will be presented.