A1681
Title: Multilevel joint model for correlated survival and binary outcome data: Application to a multicentre trial
Authors: Richard Tawiah - Monash University (Australia) [presenting]
Abstract: In many individually randomised trials, randomisation is stratified by centre; appropriate methods of analysis require centre to be accounted for in the analysis of outcomes. This is usually done by including random intercepts for centres in outcome regression models. Moreover, in most situations, multiple outcomes that are not commensurate in scale, such as time-to-event (e.g., survival) and binary or continuous endpoints (e.g., biomarker), are measured on the same subjects, thus additionally introducing intra-subject correlation between the outcomes. Joint models remain the centrepiece for simultaneous modelling of outcomes to account for their correlation. However, joint models allowing centre-specific random effects have rarely been considered, possibly due to estimation and computational complexities. A multilevel joint model is described, presenting a unified framework that simultaneously models survival and binary outcomes, utilising bivariate nested random intercepts to characterise intra-subject correlation and centre effect. Estimation is facilitated by a semiparametric restricted maximum likelihood (REML) method. Estimation difficulties are discussed, showcasing how the method simplifies calculations. Using simulation studies and the allogeneic bone marrow transplantation trial, an individually randomised multicentre trial, the method is shown to improve statistical inference as opposed to methods that provide separate analysis for the outcomes.
